void pmap_pcid_activate(pmap_t tpmap, int ccpu) {
pcid_t new_pcid = tpmap->pmap_pcid_cpus[ccpu];
pmap_t last_pmap;
boolean_t pcid_conflict = FALSE, pending_flush = FALSE;
pmap_assert(cpu_datap(ccpu)->cpu_pmap_pcid_enabled);
if (__improbable(new_pcid == PMAP_PCID_INVALID_PCID)) {
new_pcid = tpmap->pmap_pcid_cpus[ccpu] = pmap_pcid_allocate_pcid(ccpu);
}
pmap_assert(new_pcid != PMAP_PCID_INVALID_PCID);
#ifdef PCID_ASSERT
cpu_datap(ccpu)->cpu_last_pcid = cpu_datap(ccpu)->cpu_active_pcid;
#endif
cpu_datap(ccpu)->cpu_active_pcid = new_pcid;
pending_flush = (tpmap->pmap_pcid_coherency_vector[ccpu] != 0);
if (__probable(pending_flush == FALSE)) {
last_pmap = cpu_datap(ccpu)->cpu_pcid_last_pmap_dispatched[new_pcid];
pcid_conflict = ((last_pmap != NULL) &&(tpmap != last_pmap));
}
if (__improbable(pending_flush || pcid_conflict)) {
pmap_pcid_validate_cpu(tpmap, ccpu);
}
/* Consider making this a unique id */
cpu_datap(ccpu)->cpu_pcid_last_pmap_dispatched[new_pcid] = tpmap;
pmap_assert(new_pcid < PMAP_PCID_MAX_PCID);
pmap_assert(((tpmap == kernel_pmap) && new_pcid == 0) || ((new_pcid != PMAP_PCID_INVALID_PCID) && (new_pcid != 0)));
#if PMAP_ASSERT
pcid_record_array[ccpu % PCID_RECORD_SIZE] = tpmap->pm_cr3 | new_pcid | (((uint64_t)(!(pending_flush || pcid_conflict))) <<63);
pml4_entry_t *pml4 = pmap64_pml4(tpmap, 0ULL);
/* Diagnostic to detect pagetable anchor corruption */
if (pml4[KERNEL_PML4_INDEX] != kernel_pmap->pm_pml4[KERNEL_PML4_INDEX])
__asm__ volatile("int3");
#endif /* PMAP_ASSERT */
set_cr3_composed(tpmap->pm_cr3, new_pcid, !(pending_flush || pcid_conflict));
if (!pending_flush) {
/* We did not previously observe a pending invalidation for this
* ASID. However, the load from the coherency vector
* could've been reordered ahead of the store to the
* active_cr3 field (in the context switch path, our
* caller). Re-consult the pending invalidation vector
* after the CR3 write. We rely on MOV CR3's documented
* serializing property to avoid insertion of an expensive
* barrier. (DRK)
*/
pending_flush = (tpmap->pmap_pcid_coherency_vector[ccpu] != 0);
if (__improbable(pending_flush != 0)) {
pmap_pcid_validate_cpu(tpmap, ccpu);
set_cr3_composed(tpmap->pm_cr3, new_pcid, FALSE);
}
}
cpu_datap(ccpu)->cpu_pmap_pcid_coherentp = &(tpmap->pmap_pcid_coherency_vector[ccpu]);
#if DEBUG
KERNEL_DEBUG_CONSTANT(0x9c1d0000, tpmap, new_pcid, pending_flush, pcid_conflict, 0);
#endif
}
static unsigned int NOINLINE
hw_lock_lock_contended(hw_lock_t lock, uintptr_t data, uint64_t timeout, boolean_t do_panic)
{
uint64_t end = 0;
uintptr_t holder = lock->lock_data;
int i;
if (timeout == 0)
timeout = LOCK_PANIC_TIMEOUT;
#if CONFIG_DTRACE
uint64_t begin;
boolean_t dtrace_enabled = lockstat_probemap[LS_LCK_SPIN_LOCK_SPIN] != 0;
if (__improbable(dtrace_enabled))
begin = mach_absolute_time();
#endif
for ( ; ; ) {
for (i = 0; i < LOCK_SNOOP_SPINS; i++) {
cpu_pause();
#if (!__ARM_ENABLE_WFE_) || (LOCK_PRETEST)
holder = ordered_load_hw(lock);
if (holder != 0)
continue;
#endif
if (atomic_compare_exchange(&lock->lock_data, 0, data,
memory_order_acquire_smp, TRUE)) {
#if CONFIG_DTRACE
if (__improbable(dtrace_enabled)) {
uint64_t spintime = mach_absolute_time() - begin;
if (spintime > dtrace_spin_threshold)
LOCKSTAT_RECORD2(LS_LCK_SPIN_LOCK_SPIN, lock, spintime, dtrace_spin_threshold);
}
#endif
return 1;
}
}
if (end == 0) {
end = ml_get_timebase() + timeout;
}
else if (ml_get_timebase() >= end)
break;
}
if (do_panic) {
// Capture the actual time spent blocked, which may be higher than the timeout
// if a misbehaving interrupt stole this thread's CPU time.
panic("Spinlock timeout after %llu ticks, %p = %lx",
(ml_get_timebase() - end + timeout), lock, holder);
}
return 0;
}
size_t
__strlcpy_chk(char *dst, char const *src, size_t s, size_t chk_size)
{
if (__improbable(chk_size < s))
panic("__strlcpy_chk object size check failed: dst %p, src %p, (%zu < %zu)", dst, src, chk_size, s);
return strlcpy(dst, src, s);
}
void *
__memset_chk(void *dst, int c, size_t s, size_t chk_size)
{
if (__improbable(chk_size < s))
panic("__memset_chk object size check failed: dst %p, c %c, (%zu < %zu)", dst, c, chk_size, s);
return memset(dst, c, s);
}
void *
__memmove_chk(void *dst, void const *src, size_t s, size_t chk_size)
{
if (__improbable(chk_size < s))
panic("__memmove_chk object size check failed: dst %p, src %p, (%zu < %zu)", dst, src, chk_size, s);
return memmove(dst, src, s);
}
void vm_map_store_copy_insert( vm_map_t map, vm_map_entry_t after_where, vm_map_copy_t copy)
{
if (__improbable(vm_debug_events)) {
vm_map_entry_t entry;
for (entry = vm_map_copy_first_entry(copy); entry != vm_map_copy_to_entry(copy); entry = entry->vme_next) {
DTRACE_VM4(map_entry_link_copy, vm_map_t, map, vm_map_entry_t, entry, vm_address_t, entry->links.start, vm_address_t, entry->links.end);
}
}
if (map->holelistenabled) {
vm_map_entry_t entry = NULL;
entry = vm_map_copy_first_entry(copy);
while (entry != vm_map_copy_to_entry(copy)) {
vm_map_store_update_first_free(map, entry, TRUE);
entry = entry->vme_next;
}
}
vm_map_store_copy_insert_ll(map, after_where, copy);
#ifdef VM_MAP_STORE_USE_RB
if (vm_map_store_has_RB_support( &map->hdr )) {
vm_map_store_copy_insert_rb(map, after_where, copy);
}
#endif
}
void
thread_tell_urgency(int urgency,
uint64_t rt_period,
uint64_t rt_deadline,
thread_t nthread)
{
uint64_t urgency_notification_time_start, delta;
boolean_t urgency_assert = (urgency_notification_assert_abstime_threshold != 0);
assert(get_preemption_level() > 0 || ml_get_interrupts_enabled() == FALSE);
#if DEBUG
urgency_stats[cpu_number() % 64][urgency]++;
#endif
if (!pmInitDone
|| pmDispatch == NULL
|| pmDispatch->pmThreadTellUrgency == NULL)
return;
KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED,MACH_URGENCY) | DBG_FUNC_START, urgency, rt_period, rt_deadline, 0, 0);
if (__improbable((urgency_assert == TRUE)))
urgency_notification_time_start = mach_absolute_time();
current_cpu_datap()->cpu_nthread = nthread;
pmDispatch->pmThreadTellUrgency(urgency, rt_period, rt_deadline);
if (__improbable((urgency_assert == TRUE))) {
delta = mach_absolute_time() - urgency_notification_time_start;
if (__improbable(delta > urgency_notification_max_recorded)) {
/* This is not synchronized, but it doesn't matter
* if we (rarely) miss an event, as it is statistically
* unlikely that it will never recur.
*/
urgency_notification_max_recorded = delta;
if (__improbable((delta > urgency_notification_assert_abstime_threshold) && !machine_timeout_suspended()))
panic("Urgency notification callout %p exceeded threshold, 0x%llx abstime units", pmDispatch->pmThreadTellUrgency, delta);
}
}
KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED,MACH_URGENCY) | DBG_FUNC_END, urgency, rt_period, rt_deadline, 0, 0);
}
pcid_t pcid_for_pmap_cpu_tuple(pmap_t cpmap, thread_t cthread, int ccpu) {
pmap_t active_pmap = cpmap;
if (__improbable(cpmap->pagezero_accessible)) {
if ((cthread->machine.specFlags & CopyIOActive) == 0) {
active_pmap = kernel_pmap;
}
}
return active_pmap->pmap_pcid_cpus[ccpu];
}
void
_vm_map_store_entry_unlink( struct vm_map_header * mapHdr, vm_map_entry_t entry)
{
if (__improbable(vm_debug_events))
DTRACE_VM4(map_entry_unlink, vm_map_t, (char *)mapHdr - sizeof (lck_rw_t), vm_map_entry_t, entry, vm_address_t, entry->links.start, vm_address_t, entry->links.end);
vm_map_store_entry_unlink_ll(mapHdr, entry);
#ifdef VM_MAP_STORE_USE_RB
if (vm_map_store_has_RB_support( mapHdr )) {
vm_map_store_entry_unlink_rb(mapHdr, entry);
}
#endif
}
static boolean_t
timer_call_enter_internal(
timer_call_t call,
timer_call_param_t param1,
uint64_t deadline,
uint32_t flags)
{
mpqueue_head_t *queue;
mpqueue_head_t *old_queue;
spl_t s;
uint64_t slop = 0;
s = splclock();
call->soft_deadline = deadline;
call->flags = flags;
if ((flags & TIMER_CALL_CRITICAL) == 0 &&
mach_timer_coalescing_enabled) {
slop = timer_call_slop(deadline);
deadline += slop;
}
#if defined(__i386__) || defined(__x86_64__)
uint64_t ctime = mach_absolute_time();
if (__improbable(deadline < ctime)) {
uint64_t delta = (ctime - deadline);
past_deadline_timers++;
past_deadline_deltas += delta;
if (delta > past_deadline_longest)
past_deadline_longest = deadline;
if (delta < past_deadline_shortest)
past_deadline_shortest = delta;
deadline = ctime + past_deadline_timer_adjustment;
call->soft_deadline = deadline;
}
#endif
queue = timer_queue_assign(deadline);
old_queue = timer_call_enqueue_deadline_unlocked(call, queue, deadline);
CE(call)->param1 = param1;
splx(s);
return (old_queue != NULL);
}
void
_vm_map_store_entry_link( struct vm_map_header * mapHdr, vm_map_entry_t after_where, vm_map_entry_t entry)
{
assert(entry->vme_start < entry->vme_end);
if (__improbable(vm_debug_events))
DTRACE_VM4(map_entry_link, vm_map_t, (char *)mapHdr - sizeof (lck_rw_t), vm_map_entry_t, entry, vm_address_t, entry->links.start, vm_address_t, entry->links.end);
vm_map_store_entry_link_ll(mapHdr, after_where, entry);
#ifdef VM_MAP_STORE_USE_RB
if (vm_map_store_has_RB_support( mapHdr )) {
vm_map_store_entry_link_rb(mapHdr, after_where, entry);
}
#endif
#if MAP_ENTRY_INSERTION_DEBUG
fastbacktrace(&entry->vme_insertion_bt[0],
(sizeof (entry->vme_insertion_bt) / sizeof (uintptr_t)));
#endif
}
int
hfs_vnop_lookup(struct vnop_lookup_args *ap)
{
struct vnode *dvp = ap->a_dvp;
struct vnode *vp;
struct cnode *cp;
struct cnode *dcp;
struct hfsmount *hfsmp;
int error;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
struct proc *p = vfs_context_proc(ap->a_context);
int flags = cnp->cn_flags;
int force_casesensitive_lookup = proc_is_forcing_hfs_case_sensitivity(p);
int cnode_locked;
*vpp = NULL;
dcp = VTOC(dvp);
hfsmp = VTOHFS(dvp);
/*
* Lookup an entry in the cache
*
* If the lookup succeeds, the vnode is returned in *vpp,
* and a status of -1 is returned.
*
* If the lookup determines that the name does not exist
* (negative cacheing), a status of ENOENT is returned.
*
* If the lookup fails, a status of zero is returned.
*/
error = cache_lookup(dvp, vpp, cnp);
if (error != -1) {
if ((error == ENOENT) && (cnp->cn_nameiop != CREATE))
goto exit; /* found a negative cache entry */
goto lookup; /* did not find it in the cache */
}
/*
* We have a name that matched
* cache_lookup returns the vp with an iocount reference already taken
*/
error = 0;
vp = *vpp;
cp = VTOC(vp);
/* We aren't allowed to vend out vp's via lookup to the hidden directory */
if (cp->c_cnid == hfsmp->hfs_private_desc[FILE_HARDLINKS].cd_cnid ||
cp->c_cnid == hfsmp->hfs_private_desc[DIR_HARDLINKS].cd_cnid) {
/* Drop the iocount from cache_lookup */
vnode_put (vp);
error = ENOENT;
goto exit;
}
/*
* If this is a hard-link vnode then we need to update
* the name (of the link), the parent ID, the cnid, the
* text encoding and the catalog hint. This enables
* getattrlist calls to return the correct link info.
*/
/*
* Alternatively, if we are forcing a case-sensitive lookup
* on a case-insensitive volume, the namecache entry
* may have been for an incorrect case. Since we cannot
* determine case vs. normalization, redrive the catalog
* lookup based on any byte mismatch.
*/
if (((flags & ISLASTCN) && (cp->c_flag & C_HARDLINK))
|| (force_casesensitive_lookup && !(hfsmp->hfs_flags & HFS_CASE_SENSITIVE))) {
int stale_link = 0;
hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);
if ((cp->c_parentcnid != dcp->c_cnid) ||
(cnp->cn_namelen != cp->c_desc.cd_namelen) ||
(bcmp(cnp->cn_nameptr, cp->c_desc.cd_nameptr, cp->c_desc.cd_namelen) != 0)) {
struct cat_desc desc;
struct cat_attr lookup_attr;
int lockflags;
if (force_casesensitive_lookup && !(hfsmp->hfs_flags & HFS_CASE_SENSITIVE)) {
/*
* Since the name in the cnode doesn't match our lookup
* string exactly, do a full lookup.
*/
hfs_unlock (cp);
vnode_put(vp);
goto lookup;
}
/*
* Get an updated descriptor
*/
desc.cd_nameptr = (const u_int8_t *)cnp->cn_nameptr;
desc.cd_namelen = cnp->cn_namelen;
desc.cd_parentcnid = dcp->c_fileid;
desc.cd_hint = dcp->c_childhint;
desc.cd_encoding = 0;
desc.cd_cnid = 0;
desc.cd_flags = S_ISDIR(cp->c_mode) ? CD_ISDIR : 0;
/*
* Because lookups call replace_desc to put a new descriptor in
* the cnode we are modifying it is possible that this cnode's
* descriptor is out of date for the parent ID / name that
* we are trying to look up. (It may point to a different hardlink).
*
* We need to be cautious that when re-supplying the
* descriptor below that the results of the catalog lookup
* still point to the same raw inode for the hardlink. This would
* not be the case if we found something in the cache above but
* the vnode it returned no longer has a valid hardlink for the
* parent ID/filename combo we are requesting. (This is because
* hfs_unlink does not directly trigger namecache removal).
*
* As a result, before vending out the vnode (and replacing
* its descriptor) verify that the fileID is the same by comparing
* the in-cnode attributes vs. the one returned from the lookup call
* below. If they do not match, treat this lookup as if we never hit
* in the cache at all.
*/
lockflags = hfs_systemfile_lock(VTOHFS(dvp), SFL_CATALOG, HFS_SHARED_LOCK);
error = cat_lookup(VTOHFS(vp), &desc, 0, 0, &desc, &lookup_attr, NULL, NULL);
hfs_systemfile_unlock(VTOHFS(dvp), lockflags);
/*
* Note that cat_lookup may fail to find something with the name provided in the
* stack-based descriptor above. In that case, an ENOENT is a legitimate errno
* to be placed in error, which will get returned in the fastpath below.
*/
if (error == 0) {
if (lookup_attr.ca_fileid == cp->c_attr.ca_fileid) {
/* It still points to the right raw inode. Replacing the descriptor is fine */
replace_desc (cp, &desc);
/*
* Save the origin info for file and directory hardlinks. Directory hardlinks
* need the origin for '..' lookups, and file hardlinks need it to ensure that
* competing lookups do not cause us to vend different hardlinks than the ones requested.
* We want to restrict saving the cache entries to LOOKUP namei operations, since
* we're really doing this to protect getattr.
*/
if (cnp->cn_nameiop == LOOKUP) {
hfs_savelinkorigin(cp, dcp->c_fileid);
}
}
else {
/* If the fileID does not match then do NOT replace the descriptor! */
stale_link = 1;
}
}
}
hfs_unlock (cp);
if (stale_link) {
/*
* If we had a stale_link, then we need to pretend as though
* we never found this vnode and force a lookup through the
* traditional path. Drop the iocount acquired through
* cache_lookup above and force a cat lookup / getnewvnode
*/
vnode_put(vp);
goto lookup;
}
if (error) {
/*
* If the cat_lookup failed then the caller will not expect
* a vnode with an iocount on it.
*/
vnode_put(vp);
}
}
goto exit;
lookup:
/*
* The vnode was not in the name cache or it was stale.
*
* So we need to do a real lookup.
*/
cnode_locked = 0;
error = hfs_lookup(dvp, vpp, cnp, &cnode_locked, force_casesensitive_lookup);
if (cnode_locked)
hfs_unlock(VTOC(*vpp));
exit:
{
uthread_t ut = (struct uthread *)get_bsdthread_info(current_thread());
/*
* check to see if we issued any I/O while completing this lookup and
* this thread/task is throttleable... if so, throttle now
*
* this allows us to throttle in between multiple meta data reads that
* might result due to looking up a long pathname (since we'll have to
* re-enter hfs_vnop_lookup for each component of the pathnam not in
* the VFS cache), instead of waiting until the entire path lookup has
* completed and throttling at the systemcall return
*/
if (__improbable(ut->uu_lowpri_window)) {
throttle_lowpri_io(1);
}
}
return (error);
}
void
mach_call_munger64(x86_saved_state_t *state)
{
int call_number;
int argc;
mach_call_t mach_call;
struct mach_call_args args = { 0, 0, 0, 0, 0, 0, 0, 0, 0 };
x86_saved_state64_t *regs;
#if PROC_REF_DEBUG
struct uthread *ut = get_bsdthread_info(current_thread());
uthread_reset_proc_refcount(ut);
#endif
assert(is_saved_state64(state));
regs = saved_state64(state);
call_number = (int)(regs->rax & SYSCALL_NUMBER_MASK);
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger64: code=%d(%s)\n",
call_number, mach_syscall_name_table[call_number]);
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
MACHDBG_CODE(DBG_MACH_EXCP_SC,(call_number)) | DBG_FUNC_START,
regs->rdi, regs->rsi, regs->rdx, regs->r10, 0);
if (call_number < 0 || call_number >= mach_trap_count) {
i386_exception(EXC_SYSCALL, regs->rax, 1);
/* NOTREACHED */
}
mach_call = (mach_call_t)mach_trap_table[call_number].mach_trap_function;
if (mach_call == (mach_call_t)kern_invalid) {
i386_exception(EXC_SYSCALL, regs->rax, 1);
/* NOTREACHED */
}
argc = mach_trap_table[call_number].mach_trap_arg_count;
if (argc) {
int args_in_regs = MIN(6, argc);
memcpy(&args.arg1, ®s->rdi, args_in_regs * sizeof(syscall_arg_t));
if (argc > 6) {
int copyin_count;
assert(argc <= 9);
copyin_count = (argc - 6) * (int)sizeof(syscall_arg_t);
if (copyin((user_addr_t)(regs->isf.rsp + sizeof(user_addr_t)), (char *)&args.arg7, copyin_count)) {
regs->rax = KERN_INVALID_ARGUMENT;
thread_exception_return();
/* NOTREACHED */
}
}
}
#ifdef MACH_BSD
mach_kauth_cred_uthread_update();
#endif
regs->rax = (uint64_t)mach_call((void *)&args);
DEBUG_KPRINT_SYSCALL_MACH( "mach_call_munger64: retval=0x%llx\n", regs->rax);
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
MACHDBG_CODE(DBG_MACH_EXCP_SC,(call_number)) | DBG_FUNC_END,
regs->rax, 0, 0, 0, 0);
throttle_lowpri_io(1);
#if PROC_REF_DEBUG
if (__improbable(uthread_get_proc_refcount(ut) != 0)) {
panic("system call returned with uu_proc_refcount != 0");
}
#endif
thread_exception_return();
/* NOTREACHED */
}
void
mach_call_munger(x86_saved_state_t *state)
{
int argc;
int call_number;
mach_call_t mach_call;
kern_return_t retval;
struct mach_call_args args = { 0, 0, 0, 0, 0, 0, 0, 0, 0 };
x86_saved_state32_t *regs;
#if PROC_REF_DEBUG
struct uthread *ut = get_bsdthread_info(current_thread());
uthread_reset_proc_refcount(ut);
#endif
assert(is_saved_state32(state));
regs = saved_state32(state);
call_number = -(regs->eax);
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: code=%d(%s)\n",
call_number, mach_syscall_name_table[call_number]);
#if DEBUG_TRACE
kprintf("mach_call_munger(0x%08x) code=%d\n", regs, call_number);
#endif
if (call_number < 0 || call_number >= mach_trap_count) {
i386_exception(EXC_SYSCALL, call_number, 1);
/* NOTREACHED */
}
mach_call = (mach_call_t)mach_trap_table[call_number].mach_trap_function;
if (mach_call == (mach_call_t)kern_invalid) {
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: kern_invalid 0x%x\n", regs->eax);
i386_exception(EXC_SYSCALL, call_number, 1);
/* NOTREACHED */
}
argc = mach_trap_table[call_number].mach_trap_arg_count;
if (argc) {
retval = mach_call_arg_munger32(regs->uesp, &args, &mach_trap_table[call_number]);
if (retval != KERN_SUCCESS) {
regs->eax = retval;
DEBUG_KPRINT_SYSCALL_MACH(
"mach_call_munger: retval=0x%x\n", retval);
thread_exception_return();
/* NOTREACHED */
}
}
#ifdef MACH_BSD
mach_kauth_cred_uthread_update();
#endif
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
MACHDBG_CODE(DBG_MACH_EXCP_SC, (call_number)) | DBG_FUNC_START,
args.arg1, args.arg2, args.arg3, args.arg4, 0);
retval = mach_call(&args);
DEBUG_KPRINT_SYSCALL_MACH("mach_call_munger: retval=0x%x\n", retval);
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
MACHDBG_CODE(DBG_MACH_EXCP_SC,(call_number)) | DBG_FUNC_END,
retval, 0, 0, 0, 0);
regs->eax = retval;
throttle_lowpri_io(1);
#if PROC_REF_DEBUG
if (__improbable(uthread_get_proc_refcount(ut) != 0)) {
panic("system call returned with uu_proc_refcount != 0");
}
#endif
thread_exception_return();
/* NOTREACHED */
}
void
unix_syscall64(x86_saved_state_t *state)
{
thread_t thread;
void *vt;
unsigned int code;
struct sysent *callp;
int args_in_regs;
boolean_t args_start_at_rdi;
int error;
struct proc *p;
struct uthread *uthread;
x86_saved_state64_t *regs;
assert(is_saved_state64(state));
regs = saved_state64(state);
#if DEBUG
if (regs->rax == 0x2000800)
thread_exception_return();
#endif
thread = current_thread();
uthread = get_bsdthread_info(thread);
/* Get the approriate proc; may be different from task's for vfork() */
if (__probable(!(uthread->uu_flag & UT_VFORK)))
p = (struct proc *)get_bsdtask_info(current_task());
else
p = current_proc();
/* Verify that we are not being called from a task without a proc */
if (__improbable(p == NULL)) {
regs->rax = EPERM;
regs->isf.rflags |= EFL_CF;
task_terminate_internal(current_task());
thread_exception_return();
/* NOTREACHED */
}
code = regs->rax & SYSCALL_NUMBER_MASK;
DEBUG_KPRINT_SYSCALL_UNIX(
"unix_syscall64: code=%d(%s) rip=%llx\n",
code, syscallnames[code >= NUM_SYSENT ? 63 : code], regs->isf.rip);
callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
vt = (void *)uthread->uu_arg;
if (__improbable(callp == sysent)) {
/*
* indirect system call... system call number
* passed as 'arg0'
*/
code = regs->rdi;
callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
args_start_at_rdi = FALSE;
args_in_regs = 5;
} else {
args_start_at_rdi = TRUE;
args_in_regs = 6;
}
if (callp->sy_narg != 0) {
assert(callp->sy_narg <= 8); /* size of uu_arg */
args_in_regs = MIN(args_in_regs, callp->sy_narg);
memcpy(vt, args_start_at_rdi ? ®s->rdi : ®s->rsi, args_in_regs * sizeof(syscall_arg_t));
if (code != 180) {
uint64_t *ip = (uint64_t *)vt;
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
(int)(*ip), (int)(*(ip+1)), (int)(*(ip+2)), (int)(*(ip+3)), 0);
}
if (__improbable(callp->sy_narg > args_in_regs)) {
int copyin_count;
copyin_count = (callp->sy_narg - args_in_regs) * sizeof(syscall_arg_t);
error = copyin((user_addr_t)(regs->isf.rsp + sizeof(user_addr_t)), (char *)&uthread->uu_arg[args_in_regs], copyin_count);
if (error) {
regs->rax = error;
regs->isf.rflags |= EFL_CF;
thread_exception_return();
/* NOTREACHED */
}
}
} else
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
0, 0, 0, 0, 0);
/*
* Delayed binding of thread credential to process credential, if we
* are not running with an explicitly set thread credential.
*/
kauth_cred_uthread_update(uthread, p);
uthread->uu_rval[0] = 0;
uthread->uu_rval[1] = 0;
uthread->uu_flag |= UT_NOTCANCELPT;
uthread->syscall_code = code;
#ifdef JOE_DEBUG
uthread->uu_iocount = 0;
uthread->uu_vpindex = 0;
#endif
AUDIT_SYSCALL_ENTER(code, p, uthread);
error = (*(callp->sy_call))((void *) p, vt, &(uthread->uu_rval[0]));
AUDIT_SYSCALL_EXIT(code, p, uthread, error);
#ifdef JOE_DEBUG
if (uthread->uu_iocount)
printf("system call returned with uu_iocount != 0\n");
#endif
#if CONFIG_DTRACE
uthread->t_dtrace_errno = error;
#endif /* CONFIG_DTRACE */
if (__improbable(error == ERESTART)) {
/*
* all system calls come through via the syscall instruction
* in 64 bit mode... its 2 bytes in length
* move the user's pc back to repeat the syscall:
*/
pal_syscall_restart( thread, state );
}
else if (error != EJUSTRETURN) {
if (__improbable(error)) {
regs->rax = error;
regs->isf.rflags |= EFL_CF; /* carry bit */
} else { /* (not error) */
switch (callp->sy_return_type) {
case _SYSCALL_RET_INT_T:
regs->rax = uthread->uu_rval[0];
regs->rdx = uthread->uu_rval[1];
break;
case _SYSCALL_RET_UINT_T:
regs->rax = ((u_int)uthread->uu_rval[0]);
regs->rdx = ((u_int)uthread->uu_rval[1]);
break;
case _SYSCALL_RET_OFF_T:
case _SYSCALL_RET_ADDR_T:
case _SYSCALL_RET_SIZE_T:
case _SYSCALL_RET_SSIZE_T:
case _SYSCALL_RET_UINT64_T:
regs->rax = *((uint64_t *)(&uthread->uu_rval[0]));
regs->rdx = 0;
break;
case _SYSCALL_RET_NONE:
break;
default:
panic("unix_syscall: unknown return type");
break;
}
regs->isf.rflags &= ~EFL_CF;
}
}
DEBUG_KPRINT_SYSCALL_UNIX(
"unix_syscall64: error=%d retval=(%llu,%llu)\n",
error, regs->rax, regs->rdx);
uthread->uu_flag &= ~UT_NOTCANCELPT;
if (__improbable(uthread->uu_lowpri_window)) {
/*
* task is marked as a low priority I/O type
* and the I/O we issued while in this system call
* collided with normal I/O operations... we'll
* delay in order to mitigate the impact of this
* task on the normal operation of the system
*/
throttle_lowpri_io(1);
}
if (__probable(code != 180))
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_END,
error, uthread->uu_rval[0], uthread->uu_rval[1], p->p_pid, 0);
thread_exception_return();
/* NOTREACHED */
}
/*
* Event timer interrupt.
*
* XXX a drawback of this implementation is that events serviced earlier must not set deadlines
* that occur before the entire chain completes.
*
* XXX a better implementation would use a set of generic callouts and iterate over them
*/
void
timer_intr(int user_mode,
uint64_t rip)
{
uint64_t abstime;
rtclock_timer_t *mytimer;
cpu_data_t *pp;
int64_t latency;
uint64_t pmdeadline;
boolean_t timer_processed = FALSE;
pp = current_cpu_datap();
SCHED_STATS_TIMER_POP(current_processor());
abstime = mach_absolute_time(); /* Get the time now */
/* has a pending clock timer expired? */
mytimer = &pp->rtclock_timer; /* Point to the event timer */
if ((timer_processed = ((mytimer->deadline <= abstime) ||
(abstime >= (mytimer->queue.earliest_soft_deadline))))) {
/*
* Log interrupt service latency (-ve value expected by tool)
* a non-PM event is expected next.
* The requested deadline may be earlier than when it was set
* - use MAX to avoid reporting bogus latencies.
*/
latency = (int64_t) (abstime - MAX(mytimer->deadline,
mytimer->when_set));
/* Log zero timer latencies when opportunistically processing
* coalesced timers.
*/
if (latency < 0) {
TCOAL_DEBUG(0xEEEE0000, abstime, mytimer->queue.earliest_soft_deadline, abstime - mytimer->queue.earliest_soft_deadline, 0, 0);
latency = 0;
}
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
DECR_TRAP_LATENCY | DBG_FUNC_NONE,
-latency,
((user_mode != 0) ? rip : VM_KERNEL_UNSLIDE(rip)),
user_mode, 0, 0);
mytimer->has_expired = TRUE; /* Remember that we popped */
mytimer->deadline = timer_queue_expire(&mytimer->queue, abstime);
mytimer->has_expired = FALSE;
/* Get the time again since we ran a bit */
abstime = mach_absolute_time();
mytimer->when_set = abstime;
}
/* is it time for power management state change? */
if ((pmdeadline = pmCPUGetDeadline(pp)) && (pmdeadline <= abstime)) {
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
DECR_PM_DEADLINE | DBG_FUNC_START,
0, 0, 0, 0, 0);
pmCPUDeadline(pp);
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
DECR_PM_DEADLINE | DBG_FUNC_END,
0, 0, 0, 0, 0);
timer_processed = TRUE;
}
/* schedule our next deadline */
x86_lcpu()->rtcDeadline = EndOfAllTime;
timer_resync_deadlines();
if (__improbable(timer_processed == FALSE))
spurious_timers++;
}
static boolean_t
timer_call_enter_internal(
timer_call_t call,
timer_call_param_t param1,
uint64_t deadline,
uint64_t leeway,
uint32_t flags,
boolean_t ratelimited)
{
mpqueue_head_t *queue = NULL;
mpqueue_head_t *old_queue;
spl_t s;
uint64_t slop;
uint32_t urgency;
uint64_t sdeadline, ttd;
s = splclock();
sdeadline = deadline;
uint64_t ctime = mach_absolute_time();
TIMER_KDEBUG_TRACE(KDEBUG_TRACE,
DECR_TIMER_ENTER | DBG_FUNC_START,
VM_KERNEL_UNSLIDE_OR_PERM(call),
VM_KERNEL_UNSLIDE_OR_PERM(param1), deadline, flags, 0);
urgency = (flags & TIMER_CALL_URGENCY_MASK);
boolean_t slop_ratelimited = FALSE;
slop = timer_call_slop(deadline, ctime, urgency, current_thread(), &slop_ratelimited);
if ((flags & TIMER_CALL_LEEWAY) != 0 && leeway > slop)
slop = leeway;
if (UINT64_MAX - deadline <= slop) {
deadline = UINT64_MAX;
} else {
deadline += slop;
}
if (__improbable(deadline < ctime)) {
uint64_t delta = (ctime - deadline);
past_deadline_timers++;
past_deadline_deltas += delta;
if (delta > past_deadline_longest)
past_deadline_longest = deadline;
if (delta < past_deadline_shortest)
past_deadline_shortest = delta;
deadline = ctime + past_deadline_timer_adjustment;
sdeadline = deadline;
}
if (ratelimited || slop_ratelimited) {
flags |= TIMER_CALL_RATELIMITED;
} else {
flags &= ~TIMER_CALL_RATELIMITED;
}
ttd = sdeadline - ctime;
#if CONFIG_DTRACE
DTRACE_TMR7(callout__create, timer_call_func_t, TCE(call)->func,
timer_call_param_t, TCE(call)->param0, uint32_t, flags,
(deadline - sdeadline),
(ttd >> 32), (unsigned) (ttd & 0xFFFFFFFF), call);
#endif
/* Program timer callout parameters under the appropriate per-CPU or
* longterm queue lock. The callout may have been previously enqueued
* and in-flight on this or another timer queue.
*/
if (!ratelimited && !slop_ratelimited) {
queue = timer_longterm_enqueue_unlocked(call, ctime, deadline, &old_queue, sdeadline, ttd, param1, flags);
}
if (queue == NULL) {
queue = timer_queue_assign(deadline);
old_queue = timer_call_enqueue_deadline_unlocked(call, queue, deadline, sdeadline, ttd, param1, flags);
}
#if TIMER_TRACE
TCE(call)->entry_time = ctime;
#endif
TIMER_KDEBUG_TRACE(KDEBUG_TRACE,
DECR_TIMER_ENTER | DBG_FUNC_END,
VM_KERNEL_UNSLIDE_OR_PERM(call),
(old_queue != NULL), deadline, queue->count, 0);
splx(s);
return (old_queue != NULL);
}
/*
* Function: unix_syscall
*
* Inputs: regs - pointer to i386 save area
*
* Outputs: none
*/
void
unix_syscall(x86_saved_state_t *state)
{
thread_t thread;
void *vt;
unsigned int code;
struct sysent *callp;
int error;
vm_offset_t params;
struct proc *p;
struct uthread *uthread;
x86_saved_state32_t *regs;
boolean_t is_vfork;
assert(is_saved_state32(state));
regs = saved_state32(state);
#if DEBUG
if (regs->eax == 0x800)
thread_exception_return();
#endif
thread = current_thread();
uthread = get_bsdthread_info(thread);
/* Get the approriate proc; may be different from task's for vfork() */
is_vfork = uthread->uu_flag & UT_VFORK;
if (__improbable(is_vfork != 0))
p = current_proc();
else
p = (struct proc *)get_bsdtask_info(current_task());
/* Verify that we are not being called from a task without a proc */
if (__improbable(p == NULL)) {
regs->eax = EPERM;
regs->efl |= EFL_CF;
task_terminate_internal(current_task());
thread_exception_return();
/* NOTREACHED */
}
code = regs->eax & I386_SYSCALL_NUMBER_MASK;
DEBUG_KPRINT_SYSCALL_UNIX("unix_syscall: code=%d(%s) eip=%u\n",
code, syscallnames[code >= NUM_SYSENT ? 63 : code], (uint32_t)regs->eip);
params = (vm_offset_t) (regs->uesp + sizeof (int));
regs->efl &= ~(EFL_CF);
callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
if (__improbable(callp == sysent)) {
code = fuword(params);
params += sizeof(int);
callp = (code >= NUM_SYSENT) ? &sysent[63] : &sysent[code];
}
vt = (void *)uthread->uu_arg;
if (callp->sy_arg_bytes != 0) {
#if CONFIG_REQUIRES_U32_MUNGING
sy_munge_t *mungerp;
#else
#error U32 syscalls on x86_64 kernel requires munging
#endif
uint32_t nargs;
assert((unsigned) callp->sy_arg_bytes <= sizeof (uthread->uu_arg));
nargs = callp->sy_arg_bytes;
error = copyin((user_addr_t) params, (char *) vt, nargs);
if (error) {
regs->eax = error;
regs->efl |= EFL_CF;
thread_exception_return();
/* NOTREACHED */
}
if (__probable(code != 180)) {
int *ip = (int *)vt;
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
*ip, *(ip+1), *(ip+2), *(ip+3), 0);
}
#if CONFIG_REQUIRES_U32_MUNGING
mungerp = callp->sy_arg_munge32;
if (mungerp != NULL)
(*mungerp)(vt);
#endif
} else
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_START,
0, 0, 0, 0, 0);
/*
* Delayed binding of thread credential to process credential, if we
* are not running with an explicitly set thread credential.
*/
kauth_cred_uthread_update(uthread, p);
uthread->uu_rval[0] = 0;
uthread->uu_rval[1] = 0;
uthread->uu_flag |= UT_NOTCANCELPT;
uthread->syscall_code = code;
#ifdef JOE_DEBUG
uthread->uu_iocount = 0;
uthread->uu_vpindex = 0;
#endif
AUDIT_SYSCALL_ENTER(code, p, uthread);
error = (*(callp->sy_call))((void *) p, (void *) vt, &(uthread->uu_rval[0]));
AUDIT_SYSCALL_EXIT(code, p, uthread, error);
#ifdef JOE_DEBUG
if (uthread->uu_iocount)
printf("system call returned with uu_iocount != 0\n");
#endif
#if CONFIG_DTRACE
uthread->t_dtrace_errno = error;
#endif /* CONFIG_DTRACE */
if (__improbable(error == ERESTART)) {
/*
* Move the user's pc back to repeat the syscall:
* 5 bytes for a sysenter, or 2 for an int 8x.
* The SYSENTER_TF_CS covers single-stepping over a sysenter
* - see debug trap handler in idt.s/idt64.s
*/
pal_syscall_restart(thread, state);
}
else if (error != EJUSTRETURN) {
if (__improbable(error)) {
regs->eax = error;
regs->efl |= EFL_CF; /* carry bit */
} else { /* (not error) */
/*
* We split retval across two registers, in case the
* syscall had a 64-bit return value, in which case
* eax/edx matches the function call ABI.
*/
regs->eax = uthread->uu_rval[0];
regs->edx = uthread->uu_rval[1];
}
}
DEBUG_KPRINT_SYSCALL_UNIX(
"unix_syscall: error=%d retval=(%u,%u)\n",
error, regs->eax, regs->edx);
uthread->uu_flag &= ~UT_NOTCANCELPT;
if (__improbable(uthread->uu_lowpri_window)) {
/*
* task is marked as a low priority I/O type
* and the I/O we issued while in this system call
* collided with normal I/O operations... we'll
* delay in order to mitigate the impact of this
* task on the normal operation of the system
*/
throttle_lowpri_io(1);
}
if (__probable(code != 180))
KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
BSDDBG_CODE(DBG_BSD_EXCP_SC, code) | DBG_FUNC_END,
error, uthread->uu_rval[0], uthread->uu_rval[1], p->p_pid, 0);
if (__improbable(!is_vfork && callp->sy_call == (sy_call_t *)execve && !error)) {
pal_execve_return(thread);
}
thread_exception_return();
/* NOTREACHED */
}
vm_offset_t
gzalloc_alloc(zone_t zone, boolean_t canblock) {
vm_offset_t addr = 0;
if (__improbable(gzalloc_mode &&
(((zone->elem_size >= gzalloc_min) &&
(zone->elem_size <= gzalloc_max))) &&
(zone->gzalloc_exempt == 0))) {
if (get_preemption_level() != 0) {
if (canblock == TRUE) {
pdzalloc_count++;
}
else
return 0;
}
vm_offset_t rounded_size = round_page(zone->elem_size + GZHEADER_SIZE);
vm_offset_t residue = rounded_size - zone->elem_size;
vm_offset_t gzaddr = 0;
gzhdr_t *gzh;
if (!kmem_ready || (vm_page_zone == ZONE_NULL)) {
/* Early allocations are supplied directly from the
* reserve.
*/
if (gzalloc_reserve_size < rounded_size)
panic("gzalloc reserve exhausted");
gzaddr = gzalloc_reserve;
/* No guard page for these early allocations, just
* waste an additional page.
*/
gzalloc_reserve += rounded_size + PAGE_SIZE;
gzalloc_reserve_size -= rounded_size + PAGE_SIZE;
OSAddAtomic64((SInt32) (rounded_size), &gzalloc_early_alloc);
}
else {
kern_return_t kr = kernel_memory_allocate(gzalloc_map,
&gzaddr, rounded_size + (1*PAGE_SIZE),
0, KMA_KOBJECT | gzalloc_guard);
if (kr != KERN_SUCCESS)
panic("gzalloc: kernel_memory_allocate for size 0x%llx failed with %d", (uint64_t)rounded_size, kr);
}
if (gzalloc_uf_mode) {
gzaddr += PAGE_SIZE;
/* The "header" becomes a "footer" in underflow
* mode.
*/
gzh = (gzhdr_t *) (gzaddr + zone->elem_size);
addr = gzaddr;
} else {
gzh = (gzhdr_t *) (gzaddr + residue - GZHEADER_SIZE);
addr = (gzaddr + residue);
}
/* Fill with a pattern on allocation to trap uninitialized
* data use. Since the element size may be "rounded up"
* by higher layers such as the kalloc layer, this may
* also identify overruns between the originally requested
* size and the rounded size via visual inspection.
* TBD: plumb through the originally requested size,
* prior to rounding by kalloc/IOMalloc etc.
* We also add a signature and the zone of origin in a header
* prefixed to the allocation.
*/
memset((void *)gzaddr, gzalloc_fill_pattern, rounded_size);
gzh->gzone = (kmem_ready && vm_page_zone) ? zone : GZDEADZONE;
gzh->gzsize = (uint32_t) zone->elem_size;
gzh->gzsig = GZALLOC_SIGNATURE;
lock_zone(zone);
zone->count++;
zone->sum_count++;
zone->cur_size += rounded_size;
unlock_zone(zone);
OSAddAtomic64((SInt32) rounded_size, &gzalloc_allocated);
OSAddAtomic64((SInt32) (rounded_size - zone->elem_size), &gzalloc_wasted);
}
return addr;
}
static boolean_t
timer_call_enter_internal(
timer_call_t call,
timer_call_param_t param1,
uint64_t deadline,
uint64_t leeway,
uint32_t flags,
boolean_t ratelimited)
{
mpqueue_head_t *queue = NULL;
mpqueue_head_t *old_queue;
spl_t s;
uint64_t slop;
uint32_t urgency;
s = splclock();
call->soft_deadline = deadline;
call->flags = flags;
uint64_t ctime = mach_absolute_time();
TIMER_KDEBUG_TRACE(KDEBUG_TRACE,
DECR_TIMER_ENTER | DBG_FUNC_START,
call,
param1, deadline, flags, 0);
urgency = (flags & TIMER_CALL_URGENCY_MASK);
boolean_t slop_ratelimited = FALSE;
slop = timer_call_slop(deadline, ctime, urgency, current_thread(), &slop_ratelimited);
if ((flags & TIMER_CALL_LEEWAY) != 0 && leeway > slop)
slop = leeway;
if (UINT64_MAX - deadline <= slop) {
deadline = UINT64_MAX;
} else {
deadline += slop;
}
if (__improbable(deadline < ctime)) {
uint64_t delta = (ctime - deadline);
past_deadline_timers++;
past_deadline_deltas += delta;
if (delta > past_deadline_longest)
past_deadline_longest = deadline;
if (delta < past_deadline_shortest)
past_deadline_shortest = delta;
deadline = ctime + past_deadline_timer_adjustment;
call->soft_deadline = deadline;
}
/* Bit 0 of the "soft" deadline indicates that
* this particular timer call requires rate-limiting
* behaviour. Maintain the invariant deadline >= soft_deadline by
* setting bit 0 of "deadline".
*/
deadline |= 1;
if (ratelimited || slop_ratelimited) {
call->soft_deadline |= 1ULL;
} else {
call->soft_deadline &= ~0x1ULL;
}
call->ttd = call->soft_deadline - ctime;
#if CONFIG_DTRACE
DTRACE_TMR7(callout__create, timer_call_func_t, CE(call)->func,
timer_call_param_t, CE(call)->param0, uint32_t, call->flags,
(deadline - call->soft_deadline),
(call->ttd >> 32), (unsigned) (call->ttd & 0xFFFFFFFF), call);
#endif
if (!ratelimited && !slop_ratelimited) {
queue = timer_longterm_enqueue_unlocked(call, ctime, deadline, &old_queue);
}
if (queue == NULL) {
queue = timer_queue_assign(deadline);
old_queue = timer_call_enqueue_deadline_unlocked(call, queue, deadline);
}
CE(call)->param1 = param1;
#if TIMER_TRACE
CE(call)->entry_time = ctime;
#endif
TIMER_KDEBUG_TRACE(KDEBUG_TRACE,
DECR_TIMER_ENTER | DBG_FUNC_END,
call,
(old_queue != NULL), call->soft_deadline, queue->count, 0);
splx(s);
return (old_queue != NULL);
}
boolean_t gzalloc_free(zone_t zone, void *addr) {
boolean_t gzfreed = FALSE;
kern_return_t kr;
if (__improbable(gzalloc_mode &&
(((zone->elem_size >= gzalloc_min) &&
(zone->elem_size <= gzalloc_max))) &&
(zone->gzalloc_exempt == 0))) {
gzhdr_t *gzh;
vm_offset_t rounded_size = round_page(zone->elem_size + GZHEADER_SIZE);
vm_offset_t residue = rounded_size - zone->elem_size;
vm_offset_t saddr;
vm_offset_t free_addr = 0;
if (gzalloc_uf_mode) {
gzh = (gzhdr_t *)((vm_offset_t)addr + zone->elem_size);
saddr = (vm_offset_t) addr - PAGE_SIZE;
} else {
gzh = (gzhdr_t *)((vm_offset_t)addr - GZHEADER_SIZE);
saddr = ((vm_offset_t)addr) - residue;
}
assert((saddr & PAGE_MASK) == 0);
if (gzalloc_consistency_checks) {
if (gzh->gzsig != GZALLOC_SIGNATURE) {
panic("GZALLOC signature mismatch for element %p, expected 0x%x, found 0x%x", addr, GZALLOC_SIGNATURE, gzh->gzsig);
}
if (gzh->gzone != zone && (gzh->gzone != GZDEADZONE))
panic("%s: Mismatched zone or under/overflow, current zone: %p, recorded zone: %p, address: %p", __FUNCTION__, zone, gzh->gzone, (void *)addr);
/* Partially redundant given the zone check, but may flag header corruption */
if (gzh->gzsize != zone->elem_size) {
panic("Mismatched zfree or under/overflow for zone %p, recorded size: 0x%x, element size: 0x%x, address: %p\n", zone, gzh->gzsize, (uint32_t) zone->elem_size, (void *)addr);
}
}
if (!kmem_ready || gzh->gzone == GZDEADZONE) {
/* For now, just leak frees of early allocations
* performed before kmem is fully configured.
* They don't seem to get freed currently;
* consider ml_static_mfree in the future.
*/
OSAddAtomic64((SInt32) (rounded_size), &gzalloc_early_free);
return TRUE;
}
if (get_preemption_level() != 0) {
pdzfree_count++;
}
if (gzfc_size) {
/* Either write protect or unmap the newly freed
* allocation
*/
kr = vm_map_protect(
gzalloc_map,
saddr,
saddr + rounded_size + (1 * PAGE_SIZE),
gzalloc_prot,
FALSE);
if (kr != KERN_SUCCESS)
panic("%s: vm_map_protect: %p, 0x%x", __FUNCTION__, (void *)saddr, kr);
} else {
free_addr = saddr;
}
lock_zone(zone);
/* Insert newly freed element into the protected free element
* cache, and rotate out the LRU element.
*/
if (gzfc_size) {
if (zone->gz.gzfc_index >= gzfc_size) {
zone->gz.gzfc_index = 0;
}
free_addr = zone->gz.gzfc[zone->gz.gzfc_index];
zone->gz.gzfc[zone->gz.gzfc_index++] = saddr;
}
if (free_addr) {
zone->count--;
zone->cur_size -= rounded_size;
}
unlock_zone(zone);
if (free_addr) {
kr = vm_map_remove(
gzalloc_map,
free_addr,
free_addr + rounded_size + (1 * PAGE_SIZE),
VM_MAP_REMOVE_KUNWIRE);
if (kr != KERN_SUCCESS)
panic("gzfree: vm_map_remove: %p, 0x%x", (void *)free_addr, kr);
OSAddAtomic64((SInt32)rounded_size, &gzalloc_freed);
OSAddAtomic64(-((SInt32) (rounded_size - zone->elem_size)), &gzalloc_wasted);
}
gzfreed = TRUE;
}
return gzfreed;
}
