]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/dev/dtrace/fasttrap.c
xnu-4570.1.46.tar.gz
[apple/xnu.git] / bsd / dev / dtrace / fasttrap.c
index f8cbeb4c12b27bc8fdcdda9678626d548270bc2a..8425b98af29af8a6e7af711e5d3c1c0dc3bb5404 100644 (file)
@@ -31,6 +31,7 @@
 #include <sys/types.h>
 #include <sys/time.h>
 
+#include <sys/codesign.h>
 #include <sys/errno.h>
 #include <sys/stat.h>
 #include <sys/conf.h>
 #include <sys/dtrace_glue.h>
 #include <sys/dtrace_ptss.h>
 
+#include <kern/cs_blobs.h>
+#include <kern/thread.h>
 #include <kern/zalloc.h>
 
+#include <mach/thread_act.h>
+
+extern kern_return_t kernel_thread_start_priority(thread_continue_t continuation, void *parameter, integer_t priority, thread_t *new_thread);
+
 /* Solaris proc_t is the struct. Darwin's proc_t is a pointer to it. */
 #define proc_t struct proc /* Steer clear of the Darwin typedef for proc_t */
 
@@ -139,9 +146,15 @@ qsort(void *a, size_t n, size_t es, int (*cmp)(const void *, const void *));
 static dev_info_t *fasttrap_devi;
 static dtrace_meta_provider_id_t fasttrap_meta_id;
 
-static thread_call_t fasttrap_timeout;
+static thread_t fasttrap_cleanup_thread;
+
 static lck_mtx_t fasttrap_cleanup_mtx;
-static uint_t fasttrap_cleanup_work;
+
+
+#define FASTTRAP_CLEANUP_PROVIDER 0x1
+#define FASTTRAP_CLEANUP_TRACEPOINT 0x2
+
+static uint32_t fasttrap_cleanup_work = 0;
 
 /*
  * Generation count on modifications to the global tracepoint lookup table.
@@ -156,6 +169,7 @@ static volatile uint64_t fasttrap_mod_gen;
  */
 
 static uint32_t fasttrap_max;
+static uint32_t fasttrap_retired;
 static uint32_t fasttrap_total;
 
 
@@ -176,9 +190,9 @@ static lck_mtx_t                    fasttrap_count_mtx;     /* lock on ref count */
 static int fasttrap_tracepoint_enable(proc_t *, fasttrap_probe_t *, uint_t);
 static void fasttrap_tracepoint_disable(proc_t *, fasttrap_probe_t *, uint_t);
 
-static fasttrap_provider_t *fasttrap_provider_lookup(pid_t, fasttrap_provider_type_t, const char *,
+static fasttrap_provider_t *fasttrap_provider_lookup(proc_t*, fasttrap_provider_type_t, const char *,
     const dtrace_pattr_t *);
-static void fasttrap_provider_retire(pid_t, const char *, int);
+static void fasttrap_provider_retire(proc_t*, const char *, int);
 static void fasttrap_provider_free(fasttrap_provider_t *);
 
 static fasttrap_proc_t *fasttrap_proc_lookup(pid_t);
@@ -312,139 +326,257 @@ fasttrap_mod_barrier(uint64_t gen)
        }
 }
 
-/*
- * This is the timeout's callback for cleaning up the providers and their
- * probes.
- */
-/*ARGSUSED*/
-static void
-fasttrap_pid_cleanup_cb(void *ignored, void* ignored2)
+static void fasttrap_pid_cleanup(uint32_t);
+
+static unsigned int
+fasttrap_pid_cleanup_providers(void)
 {
-#pragma unused(ignored, ignored2)
        fasttrap_provider_t **fpp, *fp;
        fasttrap_bucket_t *bucket;
        dtrace_provider_id_t provid;
-       unsigned int i, later = 0;
+       unsigned int later = 0, i;
 
-       static volatile int in = 0;
-       ASSERT(in == 0);
-       in = 1;
+       /*
+        * Iterate over all the providers trying to remove the marked
+        * ones. If a provider is marked but not retired, we just
+        * have to take a crack at removing it -- it's no big deal if
+        * we can't.
+        */
+       for (i = 0; i < fasttrap_provs.fth_nent; i++) {
+               bucket = &fasttrap_provs.fth_table[i];
+               lck_mtx_lock(&bucket->ftb_mtx);
+               fpp = (fasttrap_provider_t **)&bucket->ftb_data;
 
-       lck_mtx_lock(&fasttrap_cleanup_mtx);
-       while (fasttrap_cleanup_work) {
-               fasttrap_cleanup_work = 0;
-               lck_mtx_unlock(&fasttrap_cleanup_mtx);
+               while ((fp = *fpp) != NULL) {
+                       if (!fp->ftp_marked) {
+                               fpp = &fp->ftp_next;
+                               continue;
+                       }
 
-               later = 0;
+                       lck_mtx_lock(&fp->ftp_mtx);
 
-               /*
-                * Iterate over all the providers trying to remove the marked
-                * ones. If a provider is marked but not retired, we just
-                * have to take a crack at removing it -- it's no big deal if
-                * we can't.
-                */
-               for (i = 0; i < fasttrap_provs.fth_nent; i++) {
-                       bucket = &fasttrap_provs.fth_table[i];
-                       lck_mtx_lock(&bucket->ftb_mtx);
-                       fpp = (fasttrap_provider_t **)&bucket->ftb_data;
+                       /*
+                        * If this provider has consumers actively
+                        * creating probes (ftp_ccount) or is a USDT
+                        * provider (ftp_mcount), we can't unregister
+                        * or even condense.
+                        */
+                       if (fp->ftp_ccount != 0 ||
+                           fp->ftp_mcount != 0) {
+                               fp->ftp_marked = 0;
+                               lck_mtx_unlock(&fp->ftp_mtx);
+                               continue;
+                       }
 
-                       while ((fp = *fpp) != NULL) {
-                               if (!fp->ftp_marked) {
-                                       fpp = &fp->ftp_next;
-                                       continue;
-                               }
+                       if (!fp->ftp_retired || fp->ftp_rcount != 0)
+                               fp->ftp_marked = 0;
 
-                               lck_mtx_lock(&fp->ftp_mtx);
+                       lck_mtx_unlock(&fp->ftp_mtx);
 
-                               /*
-                                * If this provider has consumers actively
-                                * creating probes (ftp_ccount) or is a USDT
-                                * provider (ftp_mcount), we can't unregister
-                                * or even condense.
-                                */
-                               if (fp->ftp_ccount != 0 ||
-                                   fp->ftp_mcount != 0) {
-                                       fp->ftp_marked = 0;
-                                       lck_mtx_unlock(&fp->ftp_mtx);
-                                       continue;
-                               }
+                       /*
+                        * If we successfully unregister this
+                        * provider we can remove it from the hash
+                        * chain and free the memory. If our attempt
+                        * to unregister fails and this is a retired
+                        * provider, increment our flag to try again
+                        * pretty soon. If we've consumed more than
+                        * half of our total permitted number of
+                        * probes call dtrace_condense() to try to
+                        * clean out the unenabled probes.
+                        */
+                       provid = fp->ftp_provid;
+                       if (dtrace_unregister(provid) != 0) {
+                               if (fasttrap_total > fasttrap_max / 2)
+                                       (void) dtrace_condense(provid);
+                               later += fp->ftp_marked;
+                               fpp = &fp->ftp_next;
+                       } else {
+                               *fpp = fp->ftp_next;
+                               fasttrap_provider_free(fp);
+                       }
+               }
+               lck_mtx_unlock(&bucket->ftb_mtx);
+       }
+
+       return later;
+}
 
-                               if (!fp->ftp_retired || fp->ftp_rcount != 0)
-                                       fp->ftp_marked = 0;
+#ifdef FASTTRAP_ASYNC_REMOVE
+typedef struct fasttrap_tracepoint_spec {
+       pid_t fttps_pid;
+       user_addr_t fttps_pc;
+} fasttrap_tracepoint_spec_t;
 
-                               lck_mtx_unlock(&fp->ftp_mtx);
+static fasttrap_tracepoint_spec_t *fasttrap_retired_spec;
+static size_t fasttrap_cur_retired = 0, fasttrap_retired_size;
+static lck_mtx_t fasttrap_retired_mtx;
 
-                               /*
-                                * If we successfully unregister this
-                                * provider we can remove it from the hash
-                                * chain and free the memory. If our attempt
-                                * to unregister fails and this is a retired
-                                * provider, increment our flag to try again
-                                * pretty soon. If we've consumed more than
-                                * half of our total permitted number of
-                                * probes call dtrace_condense() to try to
-                                * clean out the unenabled probes.
-                                */
-                               provid = fp->ftp_provid;
-                               if (dtrace_unregister(provid) != 0) {
-                                       if (fasttrap_total > fasttrap_max / 2)
-                                               (void) dtrace_condense(provid);
-                                       later += fp->ftp_marked;
-                                       fpp = &fp->ftp_next;
-                               } else {
-                                       *fpp = fp->ftp_next;
-                                       fasttrap_provider_free(fp);
-                               }
+#define DEFAULT_RETIRED_SIZE 256
+
+static void
+fasttrap_tracepoint_cleanup(void)
+{
+       size_t i;
+       pid_t pid = 0;
+       user_addr_t pc;
+       proc_t *p = PROC_NULL;
+       fasttrap_tracepoint_t *tp = NULL;
+       lck_mtx_lock(&fasttrap_retired_mtx);
+       fasttrap_bucket_t *bucket;
+       for (i = 0; i < fasttrap_cur_retired; i++) {
+               pc = fasttrap_retired_spec[i].fttps_pc;
+               if (fasttrap_retired_spec[i].fttps_pid != pid) {
+                       pid = fasttrap_retired_spec[i].fttps_pid;
+                       if (p != PROC_NULL) {
+                               sprunlock(p);
+                       }
+                       if ((p = sprlock(pid)) == PROC_NULL) {
+                               pid = 0;
+                               continue;
                        }
+               }
+               bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)];
+               lck_mtx_lock(&bucket->ftb_mtx);
+               for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) {
+                       if (pid == tp->ftt_pid && pc == tp->ftt_pc &&
+                       tp->ftt_proc->ftpc_acount != 0)
+                               break;
+               }
+               /*
+                * Check that the tracepoint is not gone or has not been
+                * re-activated for another probe
+                */
+               if (tp == NULL || tp->ftt_retired == 0) {
                        lck_mtx_unlock(&bucket->ftb_mtx);
+                       continue;
                }
+               fasttrap_tracepoint_remove(p, tp);
+               lck_mtx_unlock(&bucket->ftb_mtx);
+       }
+       if (p != PROC_NULL) {
+               sprunlock(p);
+       }
 
-               lck_mtx_lock(&fasttrap_cleanup_mtx);
+       fasttrap_cur_retired = 0;
+
+       lck_mtx_unlock(&fasttrap_retired_mtx);
+}
+
+void
+fasttrap_tracepoint_retire(proc_t *p, fasttrap_tracepoint_t *tp)
+{
+       if (tp->ftt_retired)
+               return;
+       lck_mtx_lock(&fasttrap_retired_mtx);
+       fasttrap_tracepoint_spec_t *s = &fasttrap_retired_spec[fasttrap_cur_retired++];
+       s->fttps_pid = p->p_pid;
+       s->fttps_pc = tp->ftt_pc;
+
+       if (fasttrap_cur_retired == fasttrap_retired_size) {
+               fasttrap_retired_size *= 2;
+               fasttrap_tracepoint_spec_t *new_retired = kmem_zalloc(
+                                       fasttrap_retired_size *
+                                       sizeof(fasttrap_tracepoint_t*),
+                                       KM_SLEEP);
+               memcpy(new_retired, fasttrap_retired_spec, sizeof(fasttrap_tracepoint_t*) * fasttrap_retired_size);
+               kmem_free(fasttrap_retired_spec, sizeof(fasttrap_tracepoint_t*) * (fasttrap_retired_size / 2));
+               fasttrap_retired_spec = new_retired;
        }
 
-       ASSERT(fasttrap_timeout != 0);
+       lck_mtx_unlock(&fasttrap_retired_mtx);
 
-       /*
-        * APPLE NOTE: You must hold the fasttrap_cleanup_mtx to do this!
-        */
-       if (fasttrap_timeout != (thread_call_t)1)
-               thread_call_free(fasttrap_timeout);
+       tp->ftt_retired = 1;
 
-       /*
-        * If we were unable to remove a retired provider, try again after
-        * a second. This situation can occur in certain circumstances where
-        * providers cannot be unregistered even though they have no probes
-        * enabled because of an execution of dtrace -l or something similar.
-        * If the timeout has been disabled (set to 1 because we're trying
-        * to detach), we set fasttrap_cleanup_work to ensure that we'll
-        * get a chance to do that work if and when the timeout is reenabled
-        * (if detach fails).
-        */
-       if (later > 0 && fasttrap_timeout != (thread_call_t)1)
-               /* The time value passed to dtrace_timeout is in nanos */
-               fasttrap_timeout = dtrace_timeout(&fasttrap_pid_cleanup_cb, NULL, NANOSEC / SEC);
-       else if (later > 0)
-               fasttrap_cleanup_work = 1;
-       else
-               fasttrap_timeout = 0;
+       fasttrap_pid_cleanup(FASTTRAP_CLEANUP_TRACEPOINT);
+}
+#else
+void fasttrap_tracepoint_retire(proc_t *p, fasttrap_tracepoint_t *tp)
+{
+       if (tp->ftt_retired)
+               return;
+
+       fasttrap_tracepoint_remove(p, tp);
+}
+#endif
+
+static void
+fasttrap_pid_cleanup_compute_priority(void)
+{
+       if (fasttrap_total > (fasttrap_max / 100 * 90) || fasttrap_retired > fasttrap_max / 2) {
+               thread_precedence_policy_data_t precedence = {12 /* BASEPRI_PREEMPT_HIGH */};
+               thread_policy_set(fasttrap_cleanup_thread, THREAD_PRECEDENCE_POLICY, (thread_policy_t) &precedence, THREAD_PRECEDENCE_POLICY_COUNT);
+       }
+       else {
+               thread_precedence_policy_data_t precedence = {-39 /* BASEPRI_USER_INITIATED */};
+               thread_policy_set(fasttrap_cleanup_thread, THREAD_PRECEDENCE_POLICY, (thread_policy_t) &precedence, THREAD_PRECEDENCE_POLICY_COUNT);
+
+       }
+}
+
+/*
+ * This is the timeout's callback for cleaning up the providers and their
+ * probes.
+ */
+/*ARGSUSED*/
+__attribute__((noreturn))
+static void
+fasttrap_pid_cleanup_cb(void)
+{
+       uint32_t work = 0;
+       lck_mtx_lock(&fasttrap_cleanup_mtx);
+       msleep(&fasttrap_pid_cleanup_cb, &fasttrap_cleanup_mtx, PRIBIO, "fasttrap_pid_cleanup_cb", NULL);
+       while (1) {
+               unsigned int later = 0;
+
+               work = atomic_and_32(&fasttrap_cleanup_work, 0);
+               lck_mtx_unlock(&fasttrap_cleanup_mtx);
+               if (work & FASTTRAP_CLEANUP_PROVIDER) {
+                       later = fasttrap_pid_cleanup_providers();
+               }
+#ifdef FASTTRAP_ASYNC_REMOVE
+               if (work & FASTTRAP_CLEANUP_TRACEPOINT) {
+                       fasttrap_tracepoint_cleanup();
+               }
+#endif
+               lck_mtx_lock(&fasttrap_cleanup_mtx);
+
+               fasttrap_pid_cleanup_compute_priority();
+               if (!fasttrap_cleanup_work) {
+                       /*
+                        * If we were unable to remove a retired provider, try again after
+                        * a second. This situation can occur in certain circumstances where
+                        * providers cannot be unregistered even though they have no probes
+                        * enabled because of an execution of dtrace -l or something similar.
+                        * If the timeout has been disabled (set to 1 because we're trying
+                        * to detach), we set fasttrap_cleanup_work to ensure that we'll
+                        * get a chance to do that work if and when the timeout is reenabled
+                        * (if detach fails).
+                        */
+                       if (later > 0) {
+                               struct timespec t = {1, 0};
+                               msleep(&fasttrap_pid_cleanup_cb, &fasttrap_cleanup_mtx, PRIBIO, "fasttrap_pid_cleanup_cb", &t);
+                       }
+                       else
+                               msleep(&fasttrap_pid_cleanup_cb, &fasttrap_cleanup_mtx, PRIBIO, "fasttrap_pid_cleanup_cb", NULL);
+               }
+       }
 
-       lck_mtx_unlock(&fasttrap_cleanup_mtx);
-       in = 0;
 }
 
 /*
  * Activates the asynchronous cleanup mechanism.
  */
 static void
-fasttrap_pid_cleanup(void)
+fasttrap_pid_cleanup(uint32_t work)
 {
        lck_mtx_lock(&fasttrap_cleanup_mtx);
-       fasttrap_cleanup_work = 1;
-       if (fasttrap_timeout == 0)
-               fasttrap_timeout = dtrace_timeout(&fasttrap_pid_cleanup_cb, NULL, NANOSEC / MILLISEC);
+       atomic_or_32(&fasttrap_cleanup_work, work);
+       fasttrap_pid_cleanup_compute_priority();
+       wakeup(&fasttrap_pid_cleanup_cb);
        lck_mtx_unlock(&fasttrap_cleanup_mtx);
 }
 
+
 /*
  * This is called from cfork() via dtrace_fasttrap_fork(). The child
  * process's address space is a (roughly) a copy of the parent process's so
@@ -458,7 +590,7 @@ fasttrap_fork(proc_t *p, proc_t *cp)
        unsigned int i;
 
        ASSERT(current_proc() == p);
-       lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
+       LCK_MTX_ASSERT(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_OWNED);
        ASSERT(p->p_dtrace_count > 0);
        ASSERT(cp->p_dtrace_count == 0);
 
@@ -477,7 +609,7 @@ fasttrap_fork(proc_t *p, proc_t *cp)
         * because we're in fork().
         */
        if (cp != sprlock(cp->p_pid)) {
-               printf("fasttrap_fork: sprlock(%d) returned a differt proc\n", cp->p_pid);
+               printf("fasttrap_fork: sprlock(%d) returned a different proc\n", cp->p_pid);
                return;
        }
        proc_unlock(cp);
@@ -527,8 +659,8 @@ static void
 fasttrap_exec_exit(proc_t *p)
 {
        ASSERT(p == current_proc());
-       lck_mtx_assert(&p->p_mlock, LCK_MTX_ASSERT_OWNED);
-       lck_mtx_assert(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_NOTOWNED);
+       LCK_MTX_ASSERT(&p->p_mlock, LCK_MTX_ASSERT_OWNED);
+       LCK_MTX_ASSERT(&p->p_dtrace_sprlock, LCK_MTX_ASSERT_NOTOWNED);
 
 
        /* APPLE NOTE: Okay, the locking here is really odd and needs some
@@ -547,15 +679,15 @@ fasttrap_exec_exit(proc_t *p)
         * We clean up the pid provider for this process here; user-land
         * static probes are handled by the meta-provider remove entry point.
         */
-       fasttrap_provider_retire(p->p_pid, FASTTRAP_PID_NAME, 0);
+       fasttrap_provider_retire(p, FASTTRAP_PID_NAME, 0);
 
        /*
         * APPLE NOTE: We also need to remove any aliased providers.
         * XXX optimization: track which provider types are instantiated
         * and only retire as needed.
         */
-       fasttrap_provider_retire(p->p_pid, FASTTRAP_OBJC_NAME, 0);
-       fasttrap_provider_retire(p->p_pid, FASTTRAP_ONESHOT_NAME, 0);
+       fasttrap_provider_retire(p, FASTTRAP_OBJC_NAME, 0);
+       fasttrap_provider_retire(p, FASTTRAP_ONESHOT_NAME, 0);
 
        /*
         * This should be called after it is no longer possible for a user
@@ -618,6 +750,7 @@ fasttrap_tracepoint_enable(proc_t *p, fasttrap_probe_t *probe, uint_t index)
 again:
        lck_mtx_lock(&bucket->ftb_mtx);
        for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) {
+               int rc = 0;
                /*
                 * Note that it's safe to access the active count on the
                 * associated proc structure because we know that at least one
@@ -634,7 +767,10 @@ again:
                 * enabled and the trap instruction hasn't been overwritten.
                 * Since this is a little hairy, we'll punt for now.
                 */
-
+               if (!tp->ftt_installed) {
+                       if (fasttrap_tracepoint_install(p, tp) != 0)
+                               rc = FASTTRAP_ENABLE_PARTIAL;
+               }
                /*
                 * This can't be the first interested probe. We don't have
                 * to worry about another thread being in the midst of
@@ -666,6 +802,8 @@ again:
                        ASSERT(0);
                }
 
+               tp->ftt_retired = 0;
+
                lck_mtx_unlock(&bucket->ftb_mtx);
 
                if (new_tp != NULL) {
@@ -673,7 +811,7 @@ again:
                        new_tp->ftt_retids = NULL;
                }
 
-               return (0);
+               return rc;
        }
 
        /*
@@ -697,7 +835,6 @@ again:
                 */
                if (fasttrap_tracepoint_install(p, new_tp) != 0)
                        rc = FASTTRAP_ENABLE_PARTIAL;
-
                /*
                 * Increment the count of the number of tracepoints active in
                 * the victim process.
@@ -705,6 +842,7 @@ again:
                //ASSERT(p->p_proc_flag & P_PR_LOCK);
                p->p_dtrace_count++;
 
+
                return (rc);
        }
 
@@ -714,6 +852,7 @@ again:
         * Initialize the tracepoint that's been preallocated with the probe.
         */
        new_tp = probe->ftp_tps[index].fit_tp;
+       new_tp->ftt_retired = 0;
 
        ASSERT(new_tp->ftt_pid == pid);
        ASSERT(new_tp->ftt_pc == pc);
@@ -1148,8 +1287,9 @@ fasttrap_pid_disable(void *arg, dtrace_id_t id, void *parg)
                lck_mtx_unlock(&provider->ftp_mtx);
        }
 
-       if (whack)
-               fasttrap_pid_cleanup();
+       if (whack) {
+               fasttrap_pid_cleanup(FASTTRAP_CLEANUP_PROVIDER);
+       }
 
        if (!probe->ftp_enabled)
                return;
@@ -1213,6 +1353,7 @@ fasttrap_pid_destroy(void *arg, dtrace_id_t id, void *parg)
        ASSERT(fasttrap_total >= probe->ftp_ntps);
 
        atomic_add_32(&fasttrap_total, -probe->ftp_ntps);
+       atomic_add_32(&fasttrap_retired, -probe->ftp_ntps);
 
        if (probe->ftp_gen + 1 >= fasttrap_mod_gen)
                fasttrap_mod_barrier(probe->ftp_gen);
@@ -1387,19 +1528,20 @@ fasttrap_proc_release(fasttrap_proc_t *proc)
 }
 
 /*
- * Lookup a fasttrap-managed provider based on its name and associated pid.
+ * Lookup a fasttrap-managed provider based on its name and associated proc.
+ * A reference to the proc must be held for the duration of the call.
  * If the pattr argument is non-NULL, this function instantiates the provider
  * if it doesn't exist otherwise it returns NULL. The provider is returned
  * with its lock held.
  */
 static fasttrap_provider_t *
-fasttrap_provider_lookup(pid_t pid, fasttrap_provider_type_t provider_type, const char *name,
+fasttrap_provider_lookup(proc_t *p, fasttrap_provider_type_t provider_type, const char *name,
     const dtrace_pattr_t *pattr)
 {
+       pid_t pid = p->p_pid;
        fasttrap_provider_t *fp, *new_fp = NULL;
        fasttrap_bucket_t *bucket;
        char provname[DTRACE_PROVNAMELEN];
-       proc_t *p;
        cred_t *cred;
 
        ASSERT(strlen(name) < sizeof (fp->ftp_name));
@@ -1429,16 +1571,12 @@ fasttrap_provider_lookup(pid_t pid, fasttrap_provider_type_t provider_type, cons
        lck_mtx_unlock(&bucket->ftb_mtx);
 
        /*
-        * Make sure the process exists, isn't a child created as the result
+        * Make sure the process isn't a child created as the result
         * of a vfork(2), and isn't a zombie (but may be in fork).
         */
-       if ((p = proc_find(pid)) == NULL) {
-               return NULL;
-       }
        proc_lock(p);
        if (p->p_lflag & (P_LINVFORK | P_LEXIT)) {
                proc_unlock(p);
-               proc_rele(p);
                return (NULL);
        }
 
@@ -1460,11 +1598,10 @@ fasttrap_provider_lookup(pid_t pid, fasttrap_provider_type_t provider_type, cons
        cred = p->p_ucred;
        // lck_mtx_unlock(&p->p_crlock);
        proc_unlock(p);
-       proc_rele(p);
 
        new_fp = kmem_zalloc(sizeof (fasttrap_provider_t), KM_SLEEP);
        ASSERT(new_fp != NULL);
-       new_fp->ftp_pid = pid;
+       new_fp->ftp_pid = p->p_pid;
        new_fp->ftp_proc = fasttrap_proc_lookup(pid);
        new_fp->ftp_provider_type = provider_type;
 
@@ -1578,19 +1715,18 @@ fasttrap_provider_free(fasttrap_provider_t *provider)
 }
 
 static void
-fasttrap_provider_retire(pid_t pid, const char *name, int mprov)
+fasttrap_provider_retire(proc_t *p, const char *name, int mprov)
 {
        fasttrap_provider_t *fp;
        fasttrap_bucket_t *bucket;
        dtrace_provider_id_t provid;
-
        ASSERT(strlen(name) < sizeof (fp->ftp_name));
 
-       bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)];
+       bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(p->p_pid, name)];
        lck_mtx_lock(&bucket->ftb_mtx);
 
        for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) {
-               if (fp->ftp_pid == pid && strncmp(fp->ftp_name, name, sizeof(fp->ftp_name)) == 0 &&
+               if (fp->ftp_pid == p->p_pid && strncmp(fp->ftp_name, name, sizeof(fp->ftp_name)) == 0 &&
                    !fp->ftp_retired)
                        break;
        }
@@ -1626,6 +1762,13 @@ fasttrap_provider_retire(pid_t pid, const char *name, int mprov)
        atomic_add_64(&fp->ftp_proc->ftpc_acount, -1);
        ASSERT(fp->ftp_proc->ftpc_acount < fp->ftp_proc->ftpc_rcount);
 
+       /*
+        * Add this provider probes to the retired count and
+        * make sure we don't add them twice
+        */
+       atomic_add_32(&fasttrap_retired, fp->ftp_pcount);
+       fp->ftp_pcount = 0;
+
        fp->ftp_retired = 1;
        fp->ftp_marked = 1;
        provid = fp->ftp_provid;
@@ -1633,14 +1776,14 @@ fasttrap_provider_retire(pid_t pid, const char *name, int mprov)
 
        /*
         * We don't have to worry about invalidating the same provider twice
-        * since fasttrap_provider_lookup() will ignore provider that have
+        * since fasttrap_provider_lookup() will ignore providers that have
         * been marked as retired.
         */
        dtrace_invalidate(provid);
 
        lck_mtx_unlock(&bucket->ftb_mtx);
 
-       fasttrap_pid_cleanup();
+       fasttrap_pid_cleanup(FASTTRAP_CLEANUP_PROVIDER);
 }
 
 static int
@@ -1658,6 +1801,7 @@ fasttrap_uint64_cmp(const void *ap, const void *bp)
 static int
 fasttrap_add_probe(fasttrap_probe_spec_t *pdata)
 {
+       proc_t *p;
        fasttrap_provider_t *provider;
        fasttrap_probe_t *pp;
        fasttrap_tracepoint_t *tp;
@@ -1702,10 +1846,27 @@ fasttrap_add_probe(fasttrap_probe_spec_t *pdata)
                        return (EINVAL);
        }
 
-       if ((provider = fasttrap_provider_lookup(pdata->ftps_pid, pdata->ftps_provider_type,
-                                                provider_name, &pid_attr)) == NULL)
+       p = proc_find(pdata->ftps_pid);
+       if (p == PROC_NULL)
                return (ESRCH);
 
+       /*
+        * Set that the process is allowed to run modified code and
+        * bail if it is not allowed to
+        */
+#if CONFIG_EMBEDDED
+       if ((p->p_csflags & (CS_KILL|CS_HARD)) && !cs_allow_invalid(p)) {
+               proc_rele(p);
+               return (EPERM);
+       }
+#endif
+       if ((provider = fasttrap_provider_lookup(p, pdata->ftps_provider_type,
+                                                provider_name, &pid_attr)) == NULL) {
+               proc_rele(p);
+               return (ESRCH);
+       }
+
+       proc_rele(p);
        /*
         * Increment this reference count to indicate that a consumer is
         * actively adding a new probe associated with this provider. This
@@ -1736,11 +1897,11 @@ fasttrap_add_probe(fasttrap_probe_spec_t *pdata)
                                continue;
 
                        atomic_add_32(&fasttrap_total, 1);
-
                        if (fasttrap_total > fasttrap_max) {
                                atomic_add_32(&fasttrap_total, -1);
                                goto no_mem;
                        }
+                       provider->ftp_pcount++;
 
                        pp = zalloc(fasttrap_probe_t_zones[1]);
                        bzero(pp, sizeof (fasttrap_probe_t));
@@ -1758,6 +1919,15 @@ fasttrap_add_probe(fasttrap_probe_spec_t *pdata)
                        tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc;
                        tp->ftt_pid = pdata->ftps_pid;
 
+#if defined(__arm__) || defined(__arm64__)
+                       /*
+                        * On arm the subinfo is used to distinguish between arm
+                        * and thumb modes.  On arm64 there is no thumb mode, so
+                        * this field is simply initialized to 0 on its way
+                        * into the kernel.
+                        */
+                       tp->ftt_fntype = pdata->ftps_arch_subinfo;
+#endif
 
                        pp->ftp_tps[0].fit_tp = tp;
                        pp->ftp_tps[0].fit_id.fti_probe = pp;
@@ -1790,7 +1960,7 @@ fasttrap_add_probe(fasttrap_probe_spec_t *pdata)
                        atomic_add_32(&fasttrap_total, -pdata->ftps_noffs);
                        goto no_mem;
                }
-
+               provider->ftp_pcount += pdata->ftps_noffs;
                ASSERT(pdata->ftps_noffs > 0);
                if (pdata->ftps_noffs < FASTTRAP_PROBE_T_ZONE_MAX_TRACEPOINTS) {
                        pp = zalloc(fasttrap_probe_t_zones[pdata->ftps_noffs]);
@@ -1812,6 +1982,16 @@ fasttrap_add_probe(fasttrap_probe_spec_t *pdata)
                        tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc;
                        tp->ftt_pid = pdata->ftps_pid;
 
+#if defined(__arm__) || defined (__arm64__)
+                       /*
+                        * On arm the subinfo is used to distinguish between arm
+                        * and thumb modes.  On arm64 there is no thumb mode, so
+                        * this field is simply initialized to 0 on its way
+                        * into the kernel.
+                        */
+                       
+                       tp->ftt_fntype = pdata->ftps_arch_subinfo;
+#endif
                        pp->ftp_tps[i].fit_tp = tp;
                        pp->ftp_tps[i].fit_id.fti_probe = pp;
                        pp->ftp_tps[i].fit_id.fti_ptype = pdata->ftps_probe_type;
@@ -1835,7 +2015,7 @@ fasttrap_add_probe(fasttrap_probe_spec_t *pdata)
        lck_mtx_unlock(&provider->ftp_mtx);
 
        if (whack)
-               fasttrap_pid_cleanup();
+               fasttrap_pid_cleanup(FASTTRAP_CLEANUP_PROVIDER);
 
        return (0);
 
@@ -1852,14 +2032,14 @@ no_mem:
        provider->ftp_marked = 1;
        lck_mtx_unlock(&provider->ftp_mtx);
 
-       fasttrap_pid_cleanup();
+       fasttrap_pid_cleanup(FASTTRAP_CLEANUP_PROVIDER);
 
        return (ENOMEM);
 }
 
 /*ARGSUSED*/
 static void *
-fasttrap_meta_provide(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid)
+fasttrap_meta_provide(void *arg, dtrace_helper_provdesc_t *dhpv, proc_t *p)
 {
 #pragma unused(arg)
        fasttrap_provider_t *provider;
@@ -1917,10 +2097,10 @@ fasttrap_meta_provide(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid)
        if (dhpv->dthpv_pattr.dtpa_args.dtat_class > DTRACE_CLASS_ISA)
                dhpv->dthpv_pattr.dtpa_args.dtat_class = DTRACE_CLASS_ISA;
 
-       if ((provider = fasttrap_provider_lookup(pid, DTFTP_PROVIDER_USDT, dhpv->dthpv_provname,
+       if ((provider = fasttrap_provider_lookup(p, DTFTP_PROVIDER_USDT, dhpv->dthpv_provname,
            &dhpv->dthpv_pattr)) == NULL) {
                cmn_err(CE_WARN, "failed to instantiate provider %s for "
-                   "process %u",  dhpv->dthpv_provname, (uint_t)pid);
+                   "process %u",  dhpv->dthpv_provname, (uint_t)p->p_pid);
                return (NULL);
        }
 
@@ -2023,6 +2203,8 @@ fasttrap_meta_create_probe(void *arg, void *parg,
                return;
        }
 
+       provider->ftp_pcount += ntps;
+
        if (ntps < FASTTRAP_PROBE_T_ZONE_MAX_TRACEPOINTS) {
                pp = zalloc(fasttrap_probe_t_zones[ntps]);
                bzero(pp, offsetof(fasttrap_probe_t, ftp_tps[ntps]));
@@ -2056,6 +2238,14 @@ fasttrap_meta_create_probe(void *arg, void *parg,
                 * Both 32 & 64 bit want to go back one byte, to point at the first NOP
                 */
                tp->ftt_pc = dhpb->dthpb_base + (int64_t)dhpb->dthpb_offs[i] - 1;
+#elif defined(__arm__) || defined(__arm64__)
+               /*
+                * All ARM and ARM64 probes are zero offset. We need to zero out the
+                * thumb bit because we still support 32bit user processes.
+                * On 64bit user processes, bit zero won't be set anyway.
+                */             
+               tp->ftt_pc = (dhpb->dthpb_base + (int64_t)dhpb->dthpb_offs[i]) & ~0x1UL;
+               tp->ftt_fntype = FASTTRAP_FN_USDT;
 #else
 #error "Architecture not supported"
 #endif
@@ -2086,6 +2276,14 @@ fasttrap_meta_create_probe(void *arg, void *parg,
                 * Both 32 & 64 bit want to go forward two bytes, to point at a single byte nop.
                 */
                tp->ftt_pc = dhpb->dthpb_base + (int64_t)dhpb->dthpb_enoffs[j] + 2;
+#elif defined(__arm__) || defined(__arm64__)
+               /*
+                * All ARM and ARM64 probes are zero offset. We need to zero out the
+                * thumb bit because we still support 32bit user processes.
+                * On 64bit user processes, bit zero won't be set anyway.
+                */                             
+               tp->ftt_pc = (dhpb->dthpb_base + (int64_t)dhpb->dthpb_enoffs[j]) & ~0x1UL;
+               tp->ftt_fntype = FASTTRAP_FN_USDT;
 #else
 #error "Architecture not supported"
 #endif
@@ -2120,7 +2318,7 @@ fasttrap_meta_create_probe(void *arg, void *parg,
 
 /*ARGSUSED*/
 static void
-fasttrap_meta_remove(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid)
+fasttrap_meta_remove(void *arg, dtrace_helper_provdesc_t *dhpv, proc_t *p)
 {
 #pragma unused(arg)
        /*
@@ -2129,13 +2327,22 @@ fasttrap_meta_remove(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid)
         * provider until that count has dropped to zero. This just puts
         * the provider on death row.
         */
-       fasttrap_provider_retire(pid, dhpv->dthpv_provname, 1);
+       fasttrap_provider_retire(p, dhpv->dthpv_provname, 1);
+}
+
+static char*
+fasttrap_meta_provider_name(void *arg)
+{
+       fasttrap_provider_t *fprovider = arg;
+       dtrace_provider_t *provider = (dtrace_provider_t*)(fprovider->ftp_provid);
+       return provider->dtpv_name;
 }
 
 static dtrace_mops_t fasttrap_mops = {
        fasttrap_meta_create_probe,
        fasttrap_meta_provide,
-       fasttrap_meta_remove
+       fasttrap_meta_remove,
+       fasttrap_meta_provider_name
 };
 
 /*
@@ -2230,7 +2437,8 @@ fasttrap_ioctl(dev_t dev, u_long cmd, user_addr_t arg, int md, cred_t *cr, int *
                        if ((p = proc_find(pid)) == PROC_NULL || p->p_stat == SIDL) {
                                if (p != PROC_NULL)
                                        proc_rele(p);
-                               return (ESRCH);
+                               ret = ESRCH;
+                               goto err;
                        }
                        // proc_lock(p);
                        // FIXME! How is this done on OS X?
@@ -2344,10 +2552,23 @@ fasttrap_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
         * Yes, this is a WAG.
         */
        fasttrap_max = (sane_size >> 28) * 100000;
+
+#if CONFIG_EMBEDDED
+#if defined(__LP64__)
+       /*
+        * On embedded, the zone map does not grow with the memory size over 1GB
+        * (see osfmk/vm/vm_init.c)
+        */
+       if (fasttrap_max > 400000) {
+               fasttrap_max = 400000;
+       }
+#endif
+#endif
        if (fasttrap_max == 0)
                fasttrap_max = 50000;
 
        fasttrap_total = 0;
+       fasttrap_retired = 0;
 
        /*
         * Conjure up the tracepoints hashtable...
@@ -2549,7 +2770,23 @@ fasttrap_init( void )
                        return;
                }
 
-               gFasttrapInited = 1;            
+               /*
+                * Start the fasttrap cleanup thread
+                */
+               kern_return_t res = kernel_thread_start_priority((thread_continue_t)fasttrap_pid_cleanup_cb, NULL, 46 /* BASEPRI_BACKGROUND */, &fasttrap_cleanup_thread);
+               if (res != KERN_SUCCESS) {
+                       panic("Could not create fasttrap_cleanup_thread");
+               }
+               thread_set_thread_name(fasttrap_cleanup_thread, "dtrace_fasttrap_cleanup_thread");
+
+#ifdef FASTTRAP_ASYNC_REMOVE
+               fasttrap_retired_size = DEFAULT_RETIRED_SIZE;
+               fasttrap_retired_spec = kmem_zalloc(fasttrap_retired_size * sizeof(fasttrap_tracepoint_t*),
+                                       KM_SLEEP);
+               lck_mtx_init(&fasttrap_retired_mtx, fasttrap_lck_grp, fasttrap_lck_attr);
+#endif
+
+               gFasttrapInited = 1;
        }
 }