2 * SPDX-License-Identifier: BSD-4-Clause
5 * Bill Paul <wpaul@windriver.com>. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 * must display the following acknowledgement:
17 * This product includes software developed by Bill Paul.
18 * 4. Neither the name of the author nor the names of any co-contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
32 * THE POSSIBILITY OF SUCH DAMAGE.
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
38 #include <sys/ctype.h>
39 #include <sys/unistd.h>
40 #include <sys/param.h>
41 #include <sys/types.h>
42 #include <sys/errno.h>
43 #include <sys/systm.h>
44 #include <sys/malloc.h>
46 #include <sys/mutex.h>
48 #include <sys/callout.h>
50 #include <sys/kernel.h>
52 #include <sys/condvar.h>
53 #include <sys/kthread.h>
54 #include <sys/module.h>
56 #include <sys/sched.h>
57 #include <sys/sysctl.h>
59 #include <machine/atomic.h>
60 #include <machine/bus.h>
61 #include <machine/stdarg.h>
62 #include <machine/resource.h>
68 #include <vm/vm_param.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_extern.h>
75 #include <compat/ndis/pe_var.h>
76 #include <compat/ndis/cfg_var.h>
77 #include <compat/ndis/resource_var.h>
78 #include <compat/ndis/ntoskrnl_var.h>
79 #include <compat/ndis/hal_var.h>
80 #include <compat/ndis/ndis_var.h>
82 #ifdef NTOSKRNL_DEBUG_TIMERS
83 static int sysctl_show_timers(SYSCTL_HANDLER_ARGS);
85 SYSCTL_PROC(_debug, OID_AUTO, ntoskrnl_timers,
86 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, NULL, 0,
87 sysctl_show_timers, "I",
88 "Show ntoskrnl timer stats");
102 typedef struct kdpc_queue kdpc_queue;
106 struct thread *we_td;
109 typedef struct wb_ext wb_ext;
111 #define NTOSKRNL_TIMEOUTS 256
112 #ifdef NTOSKRNL_DEBUG_TIMERS
113 static uint64_t ntoskrnl_timer_fires;
114 static uint64_t ntoskrnl_timer_sets;
115 static uint64_t ntoskrnl_timer_reloads;
116 static uint64_t ntoskrnl_timer_cancels;
119 struct callout_entry {
120 struct callout ce_callout;
124 typedef struct callout_entry callout_entry;
126 static struct list_entry ntoskrnl_calllist;
127 static struct mtx ntoskrnl_calllock;
128 struct kuser_shared_data kuser_shared_data;
130 static struct list_entry ntoskrnl_intlist;
131 static kspin_lock ntoskrnl_intlock;
133 static uint8_t RtlEqualUnicodeString(unicode_string *,
134 unicode_string *, uint8_t);
135 static void RtlCopyString(ansi_string *, const ansi_string *);
136 static void RtlCopyUnicodeString(unicode_string *,
138 static irp *IoBuildSynchronousFsdRequest(uint32_t, device_object *,
139 void *, uint32_t, uint64_t *, nt_kevent *, io_status_block *);
140 static irp *IoBuildAsynchronousFsdRequest(uint32_t,
141 device_object *, void *, uint32_t, uint64_t *, io_status_block *);
142 static irp *IoBuildDeviceIoControlRequest(uint32_t,
143 device_object *, void *, uint32_t, void *, uint32_t,
144 uint8_t, nt_kevent *, io_status_block *);
145 static irp *IoAllocateIrp(uint8_t, uint8_t);
146 static void IoReuseIrp(irp *, uint32_t);
147 static void IoFreeIrp(irp *);
148 static void IoInitializeIrp(irp *, uint16_t, uint8_t);
149 static irp *IoMakeAssociatedIrp(irp *, uint8_t);
150 static uint32_t KeWaitForMultipleObjects(uint32_t,
151 nt_dispatch_header **, uint32_t, uint32_t, uint32_t, uint8_t,
152 int64_t *, wait_block *);
153 static void ntoskrnl_waittest(nt_dispatch_header *, uint32_t);
154 static void ntoskrnl_satisfy_wait(nt_dispatch_header *, struct thread *);
155 static void ntoskrnl_satisfy_multiple_waits(wait_block *);
156 static int ntoskrnl_is_signalled(nt_dispatch_header *, struct thread *);
157 static void ntoskrnl_insert_timer(ktimer *, int);
158 static void ntoskrnl_remove_timer(ktimer *);
159 #ifdef NTOSKRNL_DEBUG_TIMERS
160 static void ntoskrnl_show_timers(void);
162 static void ntoskrnl_timercall(void *);
163 static void ntoskrnl_dpc_thread(void *);
164 static void ntoskrnl_destroy_dpc_threads(void);
165 static void ntoskrnl_destroy_workitem_threads(void);
166 static void ntoskrnl_workitem_thread(void *);
167 static void ntoskrnl_workitem(device_object *, void *);
168 static void ntoskrnl_unicode_to_ascii(uint16_t *, char *, int);
169 static void ntoskrnl_ascii_to_unicode(char *, uint16_t *, int);
170 static uint8_t ntoskrnl_insert_dpc(list_entry *, kdpc *);
171 static void WRITE_REGISTER_USHORT(uint16_t *, uint16_t);
172 static uint16_t READ_REGISTER_USHORT(uint16_t *);
173 static void WRITE_REGISTER_ULONG(uint32_t *, uint32_t);
174 static uint32_t READ_REGISTER_ULONG(uint32_t *);
175 static void WRITE_REGISTER_UCHAR(uint8_t *, uint8_t);
176 static uint8_t READ_REGISTER_UCHAR(uint8_t *);
177 static int64_t _allmul(int64_t, int64_t);
178 static int64_t _alldiv(int64_t, int64_t);
179 static int64_t _allrem(int64_t, int64_t);
180 static int64_t _allshr(int64_t, uint8_t);
181 static int64_t _allshl(int64_t, uint8_t);
182 static uint64_t _aullmul(uint64_t, uint64_t);
183 static uint64_t _aulldiv(uint64_t, uint64_t);
184 static uint64_t _aullrem(uint64_t, uint64_t);
185 static uint64_t _aullshr(uint64_t, uint8_t);
186 static uint64_t _aullshl(uint64_t, uint8_t);
187 static slist_entry *ntoskrnl_pushsl(slist_header *, slist_entry *);
188 static void InitializeSListHead(slist_header *);
189 static slist_entry *ntoskrnl_popsl(slist_header *);
190 static void ExFreePoolWithTag(void *, uint32_t);
191 static void ExInitializePagedLookasideList(paged_lookaside_list *,
192 lookaside_alloc_func *, lookaside_free_func *,
193 uint32_t, size_t, uint32_t, uint16_t);
194 static void ExDeletePagedLookasideList(paged_lookaside_list *);
195 static void ExInitializeNPagedLookasideList(npaged_lookaside_list *,
196 lookaside_alloc_func *, lookaside_free_func *,
197 uint32_t, size_t, uint32_t, uint16_t);
198 static void ExDeleteNPagedLookasideList(npaged_lookaside_list *);
200 *ExInterlockedPushEntrySList(slist_header *,
201 slist_entry *, kspin_lock *);
203 *ExInterlockedPopEntrySList(slist_header *, kspin_lock *);
204 static uint32_t InterlockedIncrement(volatile uint32_t *);
205 static uint32_t InterlockedDecrement(volatile uint32_t *);
206 static void ExInterlockedAddLargeStatistic(uint64_t *, uint32_t);
207 static void *MmAllocateContiguousMemory(uint32_t, uint64_t);
208 static void *MmAllocateContiguousMemorySpecifyCache(uint32_t,
209 uint64_t, uint64_t, uint64_t, enum nt_caching_type);
210 static void MmFreeContiguousMemory(void *);
211 static void MmFreeContiguousMemorySpecifyCache(void *, uint32_t,
212 enum nt_caching_type);
213 static uint32_t MmSizeOfMdl(void *, size_t);
214 static void *MmMapLockedPages(mdl *, uint8_t);
215 static void *MmMapLockedPagesSpecifyCache(mdl *,
216 uint8_t, uint32_t, void *, uint32_t, uint32_t);
217 static void MmUnmapLockedPages(void *, mdl *);
218 static device_t ntoskrnl_finddev(device_t, uint64_t, struct resource **);
219 static void RtlZeroMemory(void *, size_t);
220 static void RtlSecureZeroMemory(void *, size_t);
221 static void RtlFillMemory(void *, size_t, uint8_t);
222 static void RtlMoveMemory(void *, const void *, size_t);
223 static ndis_status RtlCharToInteger(const char *, uint32_t, uint32_t *);
224 static void RtlCopyMemory(void *, const void *, size_t);
225 static size_t RtlCompareMemory(const void *, const void *, size_t);
226 static ndis_status RtlUnicodeStringToInteger(unicode_string *,
227 uint32_t, uint32_t *);
228 static int atoi (const char *);
229 static long atol (const char *);
230 static int rand(void);
231 static void srand(unsigned int);
232 static void KeQuerySystemTime(uint64_t *);
233 static uint32_t KeTickCount(void);
234 static uint8_t IoIsWdmVersionAvailable(uint8_t, uint8_t);
235 static int32_t IoOpenDeviceRegistryKey(struct device_object *, uint32_t,
237 static void ntoskrnl_thrfunc(void *);
238 static ndis_status PsCreateSystemThread(ndis_handle *,
239 uint32_t, void *, ndis_handle, void *, void *, void *);
240 static ndis_status PsTerminateSystemThread(ndis_status);
241 static ndis_status IoGetDeviceObjectPointer(unicode_string *,
242 uint32_t, void *, device_object *);
243 static ndis_status IoGetDeviceProperty(device_object *, uint32_t,
244 uint32_t, void *, uint32_t *);
245 static void KeInitializeMutex(kmutant *, uint32_t);
246 static uint32_t KeReleaseMutex(kmutant *, uint8_t);
247 static uint32_t KeReadStateMutex(kmutant *);
248 static ndis_status ObReferenceObjectByHandle(ndis_handle,
249 uint32_t, void *, uint8_t, void **, void **);
250 static void ObfDereferenceObject(void *);
251 static uint32_t ZwClose(ndis_handle);
252 static uint32_t WmiQueryTraceInformation(uint32_t, void *, uint32_t,
254 static uint32_t WmiTraceMessage(uint64_t, uint32_t, void *, uint16_t, ...);
255 static uint32_t IoWMIRegistrationControl(device_object *, uint32_t);
256 static void *ntoskrnl_memset(void *, int, size_t);
257 static void *ntoskrnl_memmove(void *, void *, size_t);
258 static void *ntoskrnl_memchr(void *, unsigned char, size_t);
259 static char *ntoskrnl_strstr(char *, char *);
260 static char *ntoskrnl_strncat(char *, char *, size_t);
261 static int ntoskrnl_toupper(int);
262 static int ntoskrnl_tolower(int);
263 static funcptr ntoskrnl_findwrap(funcptr);
264 static uint32_t DbgPrint(char *, ...);
265 static void DbgBreakPoint(void);
266 static void KeBugCheckEx(uint32_t, u_long, u_long, u_long, u_long);
267 static int32_t KeDelayExecutionThread(uint8_t, uint8_t, int64_t *);
268 static int32_t KeSetPriorityThread(struct thread *, int32_t);
269 static void dummy(void);
271 static struct mtx ntoskrnl_dispatchlock;
272 static struct mtx ntoskrnl_interlock;
273 static kspin_lock ntoskrnl_cancellock;
274 static int ntoskrnl_kth = 0;
275 static struct nt_objref_head ntoskrnl_reflist;
276 static uma_zone_t mdl_zone;
277 static uma_zone_t iw_zone;
278 static struct kdpc_queue *kq_queues;
279 static struct kdpc_queue *wq_queues;
280 static int wq_idx = 0;
285 image_patch_table *patch;
292 mtx_init(&ntoskrnl_dispatchlock,
293 "ntoskrnl dispatch lock", MTX_NDIS_LOCK, MTX_DEF|MTX_RECURSE);
294 mtx_init(&ntoskrnl_interlock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
295 KeInitializeSpinLock(&ntoskrnl_cancellock);
296 KeInitializeSpinLock(&ntoskrnl_intlock);
297 TAILQ_INIT(&ntoskrnl_reflist);
299 InitializeListHead(&ntoskrnl_calllist);
300 InitializeListHead(&ntoskrnl_intlist);
301 mtx_init(&ntoskrnl_calllock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
303 kq_queues = ExAllocatePoolWithTag(NonPagedPool,
304 #ifdef NTOSKRNL_MULTIPLE_DPCS
305 sizeof(kdpc_queue) * mp_ncpus, 0);
307 sizeof(kdpc_queue), 0);
310 if (kq_queues == NULL)
313 wq_queues = ExAllocatePoolWithTag(NonPagedPool,
314 sizeof(kdpc_queue) * WORKITEM_THREADS, 0);
316 if (wq_queues == NULL)
319 #ifdef NTOSKRNL_MULTIPLE_DPCS
320 bzero((char *)kq_queues, sizeof(kdpc_queue) * mp_ncpus);
322 bzero((char *)kq_queues, sizeof(kdpc_queue));
324 bzero((char *)wq_queues, sizeof(kdpc_queue) * WORKITEM_THREADS);
327 * Launch the DPC threads.
330 #ifdef NTOSKRNL_MULTIPLE_DPCS
331 for (i = 0; i < mp_ncpus; i++) {
333 for (i = 0; i < 1; i++) {
337 error = kproc_create(ntoskrnl_dpc_thread, kq, &p,
338 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows DPC %d", i);
340 panic("failed to launch DPC thread");
344 * Launch the workitem threads.
347 for (i = 0; i < WORKITEM_THREADS; i++) {
349 error = kproc_create(ntoskrnl_workitem_thread, kq, &p,
350 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Workitem %d", i);
352 panic("failed to launch workitem thread");
355 patch = ntoskrnl_functbl;
356 while (patch->ipt_func != NULL) {
357 windrv_wrap((funcptr)patch->ipt_func,
358 (funcptr *)&patch->ipt_wrap,
359 patch->ipt_argcnt, patch->ipt_ftype);
363 for (i = 0; i < NTOSKRNL_TIMEOUTS; i++) {
364 e = ExAllocatePoolWithTag(NonPagedPool,
365 sizeof(callout_entry), 0);
367 panic("failed to allocate timeouts");
368 mtx_lock_spin(&ntoskrnl_calllock);
369 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
370 mtx_unlock_spin(&ntoskrnl_calllock);
374 * MDLs are supposed to be variable size (they describe
375 * buffers containing some number of pages, but we don't
376 * know ahead of time how many pages that will be). But
377 * always allocating them off the heap is very slow. As
378 * a compromise, we create an MDL UMA zone big enough to
379 * handle any buffer requiring up to 16 pages, and we
380 * use those for any MDLs for buffers of 16 pages or less
381 * in size. For buffers larger than that (which we assume
382 * will be few and far between, we allocate the MDLs off
386 mdl_zone = uma_zcreate("Windows MDL", MDL_ZONE_SIZE,
387 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
389 iw_zone = uma_zcreate("Windows WorkItem", sizeof(io_workitem),
390 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
398 image_patch_table *patch;
402 patch = ntoskrnl_functbl;
403 while (patch->ipt_func != NULL) {
404 windrv_unwrap(patch->ipt_wrap);
408 /* Stop the workitem queues. */
409 ntoskrnl_destroy_workitem_threads();
410 /* Stop the DPC queues. */
411 ntoskrnl_destroy_dpc_threads();
413 ExFreePool(kq_queues);
414 ExFreePool(wq_queues);
416 uma_zdestroy(mdl_zone);
417 uma_zdestroy(iw_zone);
419 mtx_lock_spin(&ntoskrnl_calllock);
420 while(!IsListEmpty(&ntoskrnl_calllist)) {
421 l = RemoveHeadList(&ntoskrnl_calllist);
422 e = CONTAINING_RECORD(l, callout_entry, ce_list);
423 mtx_unlock_spin(&ntoskrnl_calllock);
425 mtx_lock_spin(&ntoskrnl_calllock);
427 mtx_unlock_spin(&ntoskrnl_calllock);
429 mtx_destroy(&ntoskrnl_dispatchlock);
430 mtx_destroy(&ntoskrnl_interlock);
431 mtx_destroy(&ntoskrnl_calllock);
437 * We need to be able to reference this externally from the wrapper;
438 * GCC only generates a local implementation of memset.
441 ntoskrnl_memset(buf, ch, size)
446 return (memset(buf, ch, size));
450 ntoskrnl_memmove(dst, src, size)
455 bcopy(src, dst, size);
460 ntoskrnl_memchr(void *buf, unsigned char ch, size_t len)
463 unsigned char *p = buf;
468 } while (--len != 0);
474 ntoskrnl_strstr(s, find)
480 if ((c = *find++) != 0) {
484 if ((sc = *s++) == 0)
487 } while (strncmp(s, find, len) != 0);
493 /* Taken from libc */
495 ntoskrnl_strncat(dst, src, n)
507 if ((*d = *s++) == 0)
531 RtlEqualUnicodeString(unicode_string *str1, unicode_string *str2,
532 uint8_t caseinsensitive)
536 if (str1->us_len != str2->us_len)
539 for (i = 0; i < str1->us_len; i++) {
540 if (caseinsensitive == TRUE) {
541 if (toupper((char)(str1->us_buf[i] & 0xFF)) !=
542 toupper((char)(str2->us_buf[i] & 0xFF)))
545 if (str1->us_buf[i] != str2->us_buf[i])
554 RtlCopyString(dst, src)
556 const ansi_string *src;
558 if (src != NULL && src->as_buf != NULL && dst->as_buf != NULL) {
559 dst->as_len = min(src->as_len, dst->as_maxlen);
560 memcpy(dst->as_buf, src->as_buf, dst->as_len);
561 if (dst->as_len < dst->as_maxlen)
562 dst->as_buf[dst->as_len] = 0;
568 RtlCopyUnicodeString(dest, src)
569 unicode_string *dest;
573 if (dest->us_maxlen >= src->us_len)
574 dest->us_len = src->us_len;
576 dest->us_len = dest->us_maxlen;
577 memcpy(dest->us_buf, src->us_buf, dest->us_len);
581 ntoskrnl_ascii_to_unicode(ascii, unicode, len)
590 for (i = 0; i < len; i++) {
591 *ustr = (uint16_t)ascii[i];
597 ntoskrnl_unicode_to_ascii(unicode, ascii, len)
606 for (i = 0; i < len / 2; i++) {
607 *astr = (uint8_t)unicode[i];
613 RtlUnicodeStringToAnsiString(ansi_string *dest, unicode_string *src, uint8_t allocate)
615 if (dest == NULL || src == NULL)
616 return (STATUS_INVALID_PARAMETER);
618 dest->as_len = src->us_len / 2;
619 if (dest->as_maxlen < dest->as_len)
620 dest->as_len = dest->as_maxlen;
622 if (allocate == TRUE) {
623 dest->as_buf = ExAllocatePoolWithTag(NonPagedPool,
624 (src->us_len / 2) + 1, 0);
625 if (dest->as_buf == NULL)
626 return (STATUS_INSUFFICIENT_RESOURCES);
627 dest->as_len = dest->as_maxlen = src->us_len / 2;
629 dest->as_len = src->us_len / 2; /* XXX */
630 if (dest->as_maxlen < dest->as_len)
631 dest->as_len = dest->as_maxlen;
634 ntoskrnl_unicode_to_ascii(src->us_buf, dest->as_buf,
637 return (STATUS_SUCCESS);
641 RtlAnsiStringToUnicodeString(unicode_string *dest, ansi_string *src,
644 if (dest == NULL || src == NULL)
645 return (STATUS_INVALID_PARAMETER);
647 if (allocate == TRUE) {
648 dest->us_buf = ExAllocatePoolWithTag(NonPagedPool,
650 if (dest->us_buf == NULL)
651 return (STATUS_INSUFFICIENT_RESOURCES);
652 dest->us_len = dest->us_maxlen = strlen(src->as_buf) * 2;
654 dest->us_len = src->as_len * 2; /* XXX */
655 if (dest->us_maxlen < dest->us_len)
656 dest->us_len = dest->us_maxlen;
659 ntoskrnl_ascii_to_unicode(src->as_buf, dest->us_buf,
662 return (STATUS_SUCCESS);
666 ExAllocatePoolWithTag(pooltype, len, tag)
673 buf = malloc(len, M_DEVBUF, M_NOWAIT|M_ZERO);
681 ExFreePoolWithTag(buf, tag)
696 IoAllocateDriverObjectExtension(drv, clid, extlen, ext)
702 custom_extension *ce;
704 ce = ExAllocatePoolWithTag(NonPagedPool, sizeof(custom_extension)
708 return (STATUS_INSUFFICIENT_RESOURCES);
711 InsertTailList((&drv->dro_driverext->dre_usrext), (&ce->ce_list));
713 *ext = (void *)(ce + 1);
715 return (STATUS_SUCCESS);
719 IoGetDriverObjectExtension(drv, clid)
724 custom_extension *ce;
727 * Sanity check. Our dummy bus drivers don't have
728 * any driver extensions.
731 if (drv->dro_driverext == NULL)
734 e = drv->dro_driverext->dre_usrext.nle_flink;
735 while (e != &drv->dro_driverext->dre_usrext) {
736 ce = (custom_extension *)e;
737 if (ce->ce_clid == clid)
738 return ((void *)(ce + 1));
746 IoCreateDevice(driver_object *drv, uint32_t devextlen, unicode_string *devname,
747 uint32_t devtype, uint32_t devchars, uint8_t exclusive,
748 device_object **newdev)
752 dev = ExAllocatePoolWithTag(NonPagedPool, sizeof(device_object), 0);
754 return (STATUS_INSUFFICIENT_RESOURCES);
756 dev->do_type = devtype;
757 dev->do_drvobj = drv;
758 dev->do_currirp = NULL;
762 dev->do_devext = ExAllocatePoolWithTag(NonPagedPool,
765 if (dev->do_devext == NULL) {
767 return (STATUS_INSUFFICIENT_RESOURCES);
770 bzero(dev->do_devext, devextlen);
772 dev->do_devext = NULL;
774 dev->do_size = sizeof(device_object) + devextlen;
776 dev->do_attacheddev = NULL;
777 dev->do_nextdev = NULL;
778 dev->do_devtype = devtype;
779 dev->do_stacksize = 1;
780 dev->do_alignreq = 1;
781 dev->do_characteristics = devchars;
782 dev->do_iotimer = NULL;
783 KeInitializeEvent(&dev->do_devlock, EVENT_TYPE_SYNC, TRUE);
786 * Vpd is used for disk/tape devices,
787 * but we don't support those. (Yet.)
791 dev->do_devobj_ext = ExAllocatePoolWithTag(NonPagedPool,
792 sizeof(devobj_extension), 0);
794 if (dev->do_devobj_ext == NULL) {
795 if (dev->do_devext != NULL)
796 ExFreePool(dev->do_devext);
798 return (STATUS_INSUFFICIENT_RESOURCES);
801 dev->do_devobj_ext->dve_type = 0;
802 dev->do_devobj_ext->dve_size = sizeof(devobj_extension);
803 dev->do_devobj_ext->dve_devobj = dev;
806 * Attach this device to the driver object's list
807 * of devices. Note: this is not the same as attaching
808 * the device to the device stack. The driver's AddDevice
809 * routine must explicitly call IoAddDeviceToDeviceStack()
813 if (drv->dro_devobj == NULL) {
814 drv->dro_devobj = dev;
815 dev->do_nextdev = NULL;
817 dev->do_nextdev = drv->dro_devobj;
818 drv->dro_devobj = dev;
823 return (STATUS_SUCCESS);
835 if (dev->do_devobj_ext != NULL)
836 ExFreePool(dev->do_devobj_ext);
838 if (dev->do_devext != NULL)
839 ExFreePool(dev->do_devext);
841 /* Unlink the device from the driver's device list. */
843 prev = dev->do_drvobj->dro_devobj;
845 dev->do_drvobj->dro_devobj = dev->do_nextdev;
847 while (prev->do_nextdev != dev)
848 prev = prev->do_nextdev;
849 prev->do_nextdev = dev->do_nextdev;
856 IoGetAttachedDevice(dev)
866 while (d->do_attacheddev != NULL)
867 d = d->do_attacheddev;
873 IoBuildSynchronousFsdRequest(func, dobj, buf, len, off, event, status)
880 io_status_block *status;
884 ip = IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status);
887 ip->irp_usrevent = event;
893 IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status)
899 io_status_block *status;
902 io_stack_location *sl;
904 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
908 ip->irp_usriostat = status;
909 ip->irp_tail.irp_overlay.irp_thread = NULL;
911 sl = IoGetNextIrpStackLocation(ip);
912 sl->isl_major = func;
916 sl->isl_devobj = dobj;
917 sl->isl_fileobj = NULL;
918 sl->isl_completionfunc = NULL;
920 ip->irp_userbuf = buf;
922 if (dobj->do_flags & DO_BUFFERED_IO) {
923 ip->irp_assoc.irp_sysbuf =
924 ExAllocatePoolWithTag(NonPagedPool, len, 0);
925 if (ip->irp_assoc.irp_sysbuf == NULL) {
929 bcopy(buf, ip->irp_assoc.irp_sysbuf, len);
932 if (dobj->do_flags & DO_DIRECT_IO) {
933 ip->irp_mdl = IoAllocateMdl(buf, len, FALSE, FALSE, ip);
934 if (ip->irp_mdl == NULL) {
935 if (ip->irp_assoc.irp_sysbuf != NULL)
936 ExFreePool(ip->irp_assoc.irp_sysbuf);
940 ip->irp_userbuf = NULL;
941 ip->irp_assoc.irp_sysbuf = NULL;
944 if (func == IRP_MJ_READ) {
945 sl->isl_parameters.isl_read.isl_len = len;
947 sl->isl_parameters.isl_read.isl_byteoff = *off;
949 sl->isl_parameters.isl_read.isl_byteoff = 0;
952 if (func == IRP_MJ_WRITE) {
953 sl->isl_parameters.isl_write.isl_len = len;
955 sl->isl_parameters.isl_write.isl_byteoff = *off;
957 sl->isl_parameters.isl_write.isl_byteoff = 0;
964 IoBuildDeviceIoControlRequest(uint32_t iocode, device_object *dobj, void *ibuf,
965 uint32_t ilen, void *obuf, uint32_t olen, uint8_t isinternal,
966 nt_kevent *event, io_status_block *status)
969 io_stack_location *sl;
972 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
975 ip->irp_usrevent = event;
976 ip->irp_usriostat = status;
977 ip->irp_tail.irp_overlay.irp_thread = NULL;
979 sl = IoGetNextIrpStackLocation(ip);
980 sl->isl_major = isinternal == TRUE ?
981 IRP_MJ_INTERNAL_DEVICE_CONTROL : IRP_MJ_DEVICE_CONTROL;
985 sl->isl_devobj = dobj;
986 sl->isl_fileobj = NULL;
987 sl->isl_completionfunc = NULL;
988 sl->isl_parameters.isl_ioctl.isl_iocode = iocode;
989 sl->isl_parameters.isl_ioctl.isl_ibuflen = ilen;
990 sl->isl_parameters.isl_ioctl.isl_obuflen = olen;
992 switch(IO_METHOD(iocode)) {
993 case METHOD_BUFFERED:
999 ip->irp_assoc.irp_sysbuf =
1000 ExAllocatePoolWithTag(NonPagedPool, buflen, 0);
1001 if (ip->irp_assoc.irp_sysbuf == NULL) {
1006 if (ilen && ibuf != NULL) {
1007 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1008 bzero((char *)ip->irp_assoc.irp_sysbuf + ilen,
1011 bzero(ip->irp_assoc.irp_sysbuf, ilen);
1012 ip->irp_userbuf = obuf;
1014 case METHOD_IN_DIRECT:
1015 case METHOD_OUT_DIRECT:
1016 if (ilen && ibuf != NULL) {
1017 ip->irp_assoc.irp_sysbuf =
1018 ExAllocatePoolWithTag(NonPagedPool, ilen, 0);
1019 if (ip->irp_assoc.irp_sysbuf == NULL) {
1023 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1025 if (olen && obuf != NULL) {
1026 ip->irp_mdl = IoAllocateMdl(obuf, olen,
1029 * Normally we would MmProbeAndLockPages()
1030 * here, but we don't have to in our
1035 case METHOD_NEITHER:
1036 ip->irp_userbuf = obuf;
1037 sl->isl_parameters.isl_ioctl.isl_type3ibuf = ibuf;
1044 * Ideally, we should associate this IRP with the calling
1052 IoAllocateIrp(uint8_t stsize, uint8_t chargequota)
1056 i = ExAllocatePoolWithTag(NonPagedPool, IoSizeOfIrp(stsize), 0);
1060 IoInitializeIrp(i, IoSizeOfIrp(stsize), stsize);
1066 IoMakeAssociatedIrp(irp *ip, uint8_t stsize)
1070 associrp = IoAllocateIrp(stsize, FALSE);
1071 if (associrp == NULL)
1074 mtx_lock(&ntoskrnl_dispatchlock);
1075 associrp->irp_flags |= IRP_ASSOCIATED_IRP;
1076 associrp->irp_tail.irp_overlay.irp_thread =
1077 ip->irp_tail.irp_overlay.irp_thread;
1078 associrp->irp_assoc.irp_master = ip;
1079 mtx_unlock(&ntoskrnl_dispatchlock);
1092 IoInitializeIrp(irp *io, uint16_t psize, uint8_t ssize)
1094 bzero((char *)io, IoSizeOfIrp(ssize));
1095 io->irp_size = psize;
1096 io->irp_stackcnt = ssize;
1097 io->irp_currentstackloc = ssize;
1098 InitializeListHead(&io->irp_thlist);
1099 io->irp_tail.irp_overlay.irp_csl =
1100 (io_stack_location *)(io + 1) + ssize;
1104 IoReuseIrp(ip, status)
1110 allocflags = ip->irp_allocflags;
1111 IoInitializeIrp(ip, ip->irp_size, ip->irp_stackcnt);
1112 ip->irp_iostat.isb_status = status;
1113 ip->irp_allocflags = allocflags;
1117 IoAcquireCancelSpinLock(uint8_t *irql)
1119 KeAcquireSpinLock(&ntoskrnl_cancellock, irql);
1123 IoReleaseCancelSpinLock(uint8_t irql)
1125 KeReleaseSpinLock(&ntoskrnl_cancellock, irql);
1129 IoCancelIrp(irp *ip)
1134 IoAcquireCancelSpinLock(&cancelirql);
1135 cfunc = IoSetCancelRoutine(ip, NULL);
1136 ip->irp_cancel = TRUE;
1137 if (cfunc == NULL) {
1138 IoReleaseCancelSpinLock(cancelirql);
1141 ip->irp_cancelirql = cancelirql;
1142 MSCALL2(cfunc, IoGetCurrentIrpStackLocation(ip)->isl_devobj, ip);
1143 return (uint8_t)IoSetCancelValue(ip, TRUE);
1147 IofCallDriver(dobj, ip)
1148 device_object *dobj;
1151 driver_object *drvobj;
1152 io_stack_location *sl;
1154 driver_dispatch disp;
1156 drvobj = dobj->do_drvobj;
1158 if (ip->irp_currentstackloc <= 0)
1159 panic("IoCallDriver(): out of stack locations");
1161 IoSetNextIrpStackLocation(ip);
1162 sl = IoGetCurrentIrpStackLocation(ip);
1164 sl->isl_devobj = dobj;
1166 disp = drvobj->dro_dispatch[sl->isl_major];
1167 status = MSCALL2(disp, dobj, ip);
1173 IofCompleteRequest(irp *ip, uint8_t prioboost)
1176 device_object *dobj;
1177 io_stack_location *sl;
1180 KASSERT(ip->irp_iostat.isb_status != STATUS_PENDING,
1181 ("incorrect IRP(%p) status (STATUS_PENDING)", ip));
1183 sl = IoGetCurrentIrpStackLocation(ip);
1184 IoSkipCurrentIrpStackLocation(ip);
1187 if (sl->isl_ctl & SL_PENDING_RETURNED)
1188 ip->irp_pendingreturned = TRUE;
1190 if (ip->irp_currentstackloc != (ip->irp_stackcnt + 1))
1191 dobj = IoGetCurrentIrpStackLocation(ip)->isl_devobj;
1195 if (sl->isl_completionfunc != NULL &&
1196 ((ip->irp_iostat.isb_status == STATUS_SUCCESS &&
1197 sl->isl_ctl & SL_INVOKE_ON_SUCCESS) ||
1198 (ip->irp_iostat.isb_status != STATUS_SUCCESS &&
1199 sl->isl_ctl & SL_INVOKE_ON_ERROR) ||
1200 (ip->irp_cancel == TRUE &&
1201 sl->isl_ctl & SL_INVOKE_ON_CANCEL))) {
1202 cf = sl->isl_completionfunc;
1203 status = MSCALL3(cf, dobj, ip, sl->isl_completionctx);
1204 if (status == STATUS_MORE_PROCESSING_REQUIRED)
1207 if ((ip->irp_currentstackloc <= ip->irp_stackcnt) &&
1208 (ip->irp_pendingreturned == TRUE))
1209 IoMarkIrpPending(ip);
1212 /* move to the next. */
1213 IoSkipCurrentIrpStackLocation(ip);
1215 } while (ip->irp_currentstackloc <= (ip->irp_stackcnt + 1));
1217 if (ip->irp_usriostat != NULL)
1218 *ip->irp_usriostat = ip->irp_iostat;
1219 if (ip->irp_usrevent != NULL)
1220 KeSetEvent(ip->irp_usrevent, prioboost, FALSE);
1222 /* Handle any associated IRPs. */
1224 if (ip->irp_flags & IRP_ASSOCIATED_IRP) {
1225 uint32_t masterirpcnt;
1229 masterirp = ip->irp_assoc.irp_master;
1231 InterlockedDecrement(&masterirp->irp_assoc.irp_irpcnt);
1233 while ((m = ip->irp_mdl) != NULL) {
1234 ip->irp_mdl = m->mdl_next;
1238 if (masterirpcnt == 0)
1239 IoCompleteRequest(masterirp, IO_NO_INCREMENT);
1243 /* With any luck, these conditions will never arise. */
1245 if (ip->irp_flags & IRP_PAGING_IO) {
1246 if (ip->irp_mdl != NULL)
1247 IoFreeMdl(ip->irp_mdl);
1261 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1262 l = ntoskrnl_intlist.nle_flink;
1263 while (l != &ntoskrnl_intlist) {
1264 iobj = CONTAINING_RECORD(l, kinterrupt, ki_list);
1265 claimed = MSCALL2(iobj->ki_svcfunc, iobj, iobj->ki_svcctx);
1266 if (claimed == TRUE)
1270 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1274 KeAcquireInterruptSpinLock(iobj)
1278 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1283 KeReleaseInterruptSpinLock(kinterrupt *iobj, uint8_t irql)
1285 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1289 KeSynchronizeExecution(iobj, syncfunc, syncctx)
1296 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1297 MSCALL1(syncfunc, syncctx);
1298 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1304 * IoConnectInterrupt() is passed only the interrupt vector and
1305 * irql that a device wants to use, but no device-specific tag
1306 * of any kind. This conflicts rather badly with FreeBSD's
1307 * bus_setup_intr(), which needs the device_t for the device
1308 * requesting interrupt delivery. In order to bypass this
1309 * inconsistency, we implement a second level of interrupt
1310 * dispatching on top of bus_setup_intr(). All devices use
1311 * ntoskrnl_intr() as their ISR, and any device requesting
1312 * interrupts will be registered with ntoskrnl_intr()'s interrupt
1313 * dispatch list. When an interrupt arrives, we walk the list
1314 * and invoke all the registered ISRs. This effectively makes all
1315 * interrupts shared, but it's the only way to duplicate the
1316 * semantics of IoConnectInterrupt() and IoDisconnectInterrupt() properly.
1320 IoConnectInterrupt(kinterrupt **iobj, void *svcfunc, void *svcctx,
1321 kspin_lock *lock, uint32_t vector, uint8_t irql, uint8_t syncirql,
1322 uint8_t imode, uint8_t shared, uint32_t affinity, uint8_t savefloat)
1326 *iobj = ExAllocatePoolWithTag(NonPagedPool, sizeof(kinterrupt), 0);
1328 return (STATUS_INSUFFICIENT_RESOURCES);
1330 (*iobj)->ki_svcfunc = svcfunc;
1331 (*iobj)->ki_svcctx = svcctx;
1334 KeInitializeSpinLock(&(*iobj)->ki_lock_priv);
1335 (*iobj)->ki_lock = &(*iobj)->ki_lock_priv;
1337 (*iobj)->ki_lock = lock;
1339 KeAcquireSpinLock(&ntoskrnl_intlock, &curirql);
1340 InsertHeadList((&ntoskrnl_intlist), (&(*iobj)->ki_list));
1341 KeReleaseSpinLock(&ntoskrnl_intlock, curirql);
1343 return (STATUS_SUCCESS);
1347 IoDisconnectInterrupt(iobj)
1355 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1356 RemoveEntryList((&iobj->ki_list));
1357 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1363 IoAttachDeviceToDeviceStack(src, dst)
1367 device_object *attached;
1369 mtx_lock(&ntoskrnl_dispatchlock);
1370 attached = IoGetAttachedDevice(dst);
1371 attached->do_attacheddev = src;
1372 src->do_attacheddev = NULL;
1373 src->do_stacksize = attached->do_stacksize + 1;
1374 mtx_unlock(&ntoskrnl_dispatchlock);
1380 IoDetachDevice(topdev)
1381 device_object *topdev;
1383 device_object *tail;
1385 mtx_lock(&ntoskrnl_dispatchlock);
1387 /* First, break the chain. */
1388 tail = topdev->do_attacheddev;
1390 mtx_unlock(&ntoskrnl_dispatchlock);
1393 topdev->do_attacheddev = tail->do_attacheddev;
1394 topdev->do_refcnt--;
1396 /* Now reduce the stacksize count for the takm_il objects. */
1398 tail = topdev->do_attacheddev;
1399 while (tail != NULL) {
1400 tail->do_stacksize--;
1401 tail = tail->do_attacheddev;
1404 mtx_unlock(&ntoskrnl_dispatchlock);
1408 * For the most part, an object is considered signalled if
1409 * dh_sigstate == TRUE. The exception is for mutant objects
1410 * (mutexes), where the logic works like this:
1412 * - If the thread already owns the object and sigstate is
1413 * less than or equal to 0, then the object is considered
1414 * signalled (recursive acquisition).
1415 * - If dh_sigstate == 1, the object is also considered
1420 ntoskrnl_is_signalled(obj, td)
1421 nt_dispatch_header *obj;
1426 if (obj->dh_type == DISP_TYPE_MUTANT) {
1427 km = (kmutant *)obj;
1428 if ((obj->dh_sigstate <= 0 && km->km_ownerthread == td) ||
1429 obj->dh_sigstate == 1)
1434 if (obj->dh_sigstate > 0)
1440 ntoskrnl_satisfy_wait(obj, td)
1441 nt_dispatch_header *obj;
1446 switch (obj->dh_type) {
1447 case DISP_TYPE_MUTANT:
1448 km = (struct kmutant *)obj;
1451 * If sigstate reaches 0, the mutex is now
1452 * non-signalled (the new thread owns it).
1454 if (obj->dh_sigstate == 0) {
1455 km->km_ownerthread = td;
1456 if (km->km_abandoned == TRUE)
1457 km->km_abandoned = FALSE;
1460 /* Synchronization objects get reset to unsignalled. */
1461 case DISP_TYPE_SYNCHRONIZATION_EVENT:
1462 case DISP_TYPE_SYNCHRONIZATION_TIMER:
1463 obj->dh_sigstate = 0;
1465 case DISP_TYPE_SEMAPHORE:
1474 ntoskrnl_satisfy_multiple_waits(wb)
1481 td = wb->wb_kthread;
1484 ntoskrnl_satisfy_wait(wb->wb_object, td);
1485 cur->wb_awakened = TRUE;
1487 } while (cur != wb);
1490 /* Always called with dispatcher lock held. */
1492 ntoskrnl_waittest(obj, increment)
1493 nt_dispatch_header *obj;
1496 wait_block *w, *next;
1503 * Once an object has been signalled, we walk its list of
1504 * wait blocks. If a wait block can be awakened, then satisfy
1505 * waits as necessary and wake the thread.
1507 * The rules work like this:
1509 * If a wait block is marked as WAITTYPE_ANY, then
1510 * we can satisfy the wait conditions on the current
1511 * object and wake the thread right away. Satisfying
1512 * the wait also has the effect of breaking us out
1513 * of the search loop.
1515 * If the object is marked as WAITTYLE_ALL, then the
1516 * wait block will be part of a circularly linked
1517 * list of wait blocks belonging to a waiting thread
1518 * that's sleeping in KeWaitForMultipleObjects(). In
1519 * order to wake the thread, all the objects in the
1520 * wait list must be in the signalled state. If they
1521 * are, we then satisfy all of them and wake the
1526 e = obj->dh_waitlisthead.nle_flink;
1528 while (e != &obj->dh_waitlisthead && obj->dh_sigstate > 0) {
1529 w = CONTAINING_RECORD(e, wait_block, wb_waitlist);
1533 if (w->wb_waittype == WAITTYPE_ANY) {
1535 * Thread can be awakened if
1536 * any wait is satisfied.
1538 ntoskrnl_satisfy_wait(obj, td);
1540 w->wb_awakened = TRUE;
1543 * Thread can only be woken up
1544 * if all waits are satisfied.
1545 * If the thread is waiting on multiple
1546 * objects, they should all be linked
1547 * through the wb_next pointers in the
1553 if (ntoskrnl_is_signalled(obj, td) == FALSE) {
1557 next = next->wb_next;
1559 ntoskrnl_satisfy_multiple_waits(w);
1562 if (satisfied == TRUE)
1563 cv_broadcastpri(&we->we_cv,
1564 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
1565 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
1572 * Return the number of 100 nanosecond intervals since
1573 * January 1, 1601. (?!?!)
1582 *tval = (uint64_t)ts.tv_nsec / 100 + (uint64_t)ts.tv_sec * 10000000 +
1583 11644473600 * 10000000; /* 100ns ticks from 1601 to 1970 */
1587 KeQuerySystemTime(current_time)
1588 uint64_t *current_time;
1590 ntoskrnl_time(current_time);
1597 getmicrouptime(&tv);
1602 * KeWaitForSingleObject() is a tricky beast, because it can be used
1603 * with several different object types: semaphores, timers, events,
1604 * mutexes and threads. Semaphores don't appear very often, but the
1605 * other object types are quite common. KeWaitForSingleObject() is
1606 * what's normally used to acquire a mutex, and it can be used to
1607 * wait for a thread termination.
1609 * The Windows NDIS API is implemented in terms of Windows kernel
1610 * primitives, and some of the object manipulation is duplicated in
1611 * NDIS. For example, NDIS has timers and events, which are actually
1612 * Windows kevents and ktimers. Now, you're supposed to only use the
1613 * NDIS variants of these objects within the confines of the NDIS API,
1614 * but there are some naughty developers out there who will use
1615 * KeWaitForSingleObject() on NDIS timer and event objects, so we
1616 * have to support that as well. Conseqently, our NDIS timer and event
1617 * code has to be closely tied into our ntoskrnl timer and event code,
1618 * just as it is in Windows.
1620 * KeWaitForSingleObject() may do different things for different kinds
1623 * - For events, we check if the event has been signalled. If the
1624 * event is already in the signalled state, we just return immediately,
1625 * otherwise we wait for it to be set to the signalled state by someone
1626 * else calling KeSetEvent(). Events can be either synchronization or
1627 * notification events.
1629 * - For timers, if the timer has already fired and the timer is in
1630 * the signalled state, we just return, otherwise we wait on the
1631 * timer. Unlike an event, timers get signalled automatically when
1632 * they expire rather than someone having to trip them manually.
1633 * Timers initialized with KeInitializeTimer() are always notification
1634 * events: KeInitializeTimerEx() lets you initialize a timer as
1635 * either a notification or synchronization event.
1637 * - For mutexes, we try to acquire the mutex and if we can't, we wait
1638 * on the mutex until it's available and then grab it. When a mutex is
1639 * released, it enters the signalled state, which wakes up one of the
1640 * threads waiting to acquire it. Mutexes are always synchronization
1643 * - For threads, the only thing we do is wait until the thread object
1644 * enters a signalled state, which occurs when the thread terminates.
1645 * Threads are always notification events.
1647 * A notification event wakes up all threads waiting on an object. A
1648 * synchronization event wakes up just one. Also, a synchronization event
1649 * is auto-clearing, which means we automatically set the event back to
1650 * the non-signalled state once the wakeup is done.
1654 KeWaitForSingleObject(void *arg, uint32_t reason, uint32_t mode,
1655 uint8_t alertable, int64_t *duetime)
1658 struct thread *td = curthread;
1663 nt_dispatch_header *obj;
1668 return (STATUS_INVALID_PARAMETER);
1670 mtx_lock(&ntoskrnl_dispatchlock);
1672 cv_init(&we.we_cv, "KeWFS");
1676 * Check to see if this object is already signalled,
1677 * and just return without waiting if it is.
1679 if (ntoskrnl_is_signalled(obj, td) == TRUE) {
1680 /* Sanity check the signal state value. */
1681 if (obj->dh_sigstate != INT32_MIN) {
1682 ntoskrnl_satisfy_wait(obj, curthread);
1683 mtx_unlock(&ntoskrnl_dispatchlock);
1684 return (STATUS_SUCCESS);
1687 * There's a limit to how many times we can
1688 * recursively acquire a mutant. If we hit
1689 * the limit, something is very wrong.
1691 if (obj->dh_type == DISP_TYPE_MUTANT) {
1692 mtx_unlock(&ntoskrnl_dispatchlock);
1693 panic("mutant limit exceeded");
1698 bzero((char *)&w, sizeof(wait_block));
1701 w.wb_waittype = WAITTYPE_ANY;
1704 w.wb_awakened = FALSE;
1705 w.wb_oldpri = td->td_priority;
1707 InsertTailList((&obj->dh_waitlisthead), (&w.wb_waitlist));
1710 * The timeout value is specified in 100 nanosecond units
1711 * and can be a positive or negative number. If it's positive,
1712 * then the duetime is absolute, and we need to convert it
1713 * to an absolute offset relative to now in order to use it.
1714 * If it's negative, then the duetime is relative and we
1715 * just have to convert the units.
1718 if (duetime != NULL) {
1720 tv.tv_sec = - (*duetime) / 10000000;
1721 tv.tv_usec = (- (*duetime) / 10) -
1722 (tv.tv_sec * 1000000);
1724 ntoskrnl_time(&curtime);
1725 if (*duetime < curtime)
1726 tv.tv_sec = tv.tv_usec = 0;
1728 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1729 tv.tv_usec = ((*duetime) - curtime) / 10 -
1730 (tv.tv_sec * 1000000);
1735 if (duetime == NULL)
1736 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1738 error = cv_timedwait(&we.we_cv,
1739 &ntoskrnl_dispatchlock, tvtohz(&tv));
1741 RemoveEntryList(&w.wb_waitlist);
1743 cv_destroy(&we.we_cv);
1745 /* We timed out. Leave the object alone and return status. */
1747 if (error == EWOULDBLOCK) {
1748 mtx_unlock(&ntoskrnl_dispatchlock);
1749 return (STATUS_TIMEOUT);
1752 mtx_unlock(&ntoskrnl_dispatchlock);
1754 return (STATUS_SUCCESS);
1756 return (KeWaitForMultipleObjects(1, &obj, WAITTYPE_ALL, reason,
1757 mode, alertable, duetime, &w));
1762 KeWaitForMultipleObjects(uint32_t cnt, nt_dispatch_header *obj[], uint32_t wtype,
1763 uint32_t reason, uint32_t mode, uint8_t alertable, int64_t *duetime,
1764 wait_block *wb_array)
1766 struct thread *td = curthread;
1767 wait_block *whead, *w;
1768 wait_block _wb_array[MAX_WAIT_OBJECTS];
1769 nt_dispatch_header *cur;
1771 int i, wcnt = 0, error = 0;
1773 struct timespec t1, t2;
1774 uint32_t status = STATUS_SUCCESS;
1777 if (cnt > MAX_WAIT_OBJECTS)
1778 return (STATUS_INVALID_PARAMETER);
1779 if (cnt > THREAD_WAIT_OBJECTS && wb_array == NULL)
1780 return (STATUS_INVALID_PARAMETER);
1782 mtx_lock(&ntoskrnl_dispatchlock);
1784 cv_init(&we.we_cv, "KeWFM");
1787 if (wb_array == NULL)
1792 bzero((char *)whead, sizeof(wait_block) * cnt);
1794 /* First pass: see if we can satisfy any waits immediately. */
1799 for (i = 0; i < cnt; i++) {
1800 InsertTailList((&obj[i]->dh_waitlisthead),
1803 w->wb_object = obj[i];
1804 w->wb_waittype = wtype;
1806 w->wb_awakened = FALSE;
1807 w->wb_oldpri = td->td_priority;
1811 if (ntoskrnl_is_signalled(obj[i], td)) {
1813 * There's a limit to how many times
1814 * we can recursively acquire a mutant.
1815 * If we hit the limit, something
1818 if (obj[i]->dh_sigstate == INT32_MIN &&
1819 obj[i]->dh_type == DISP_TYPE_MUTANT) {
1820 mtx_unlock(&ntoskrnl_dispatchlock);
1821 panic("mutant limit exceeded");
1825 * If this is a WAITTYPE_ANY wait, then
1826 * satisfy the waited object and exit
1830 if (wtype == WAITTYPE_ANY) {
1831 ntoskrnl_satisfy_wait(obj[i], td);
1832 status = STATUS_WAIT_0 + i;
1837 w->wb_object = NULL;
1838 RemoveEntryList(&w->wb_waitlist);
1844 * If this is a WAITTYPE_ALL wait and all objects are
1845 * already signalled, satisfy the waits and exit now.
1848 if (wtype == WAITTYPE_ALL && wcnt == 0) {
1849 for (i = 0; i < cnt; i++)
1850 ntoskrnl_satisfy_wait(obj[i], td);
1851 status = STATUS_SUCCESS;
1856 * Create a circular waitblock list. The waitcount
1857 * must always be non-zero when we get here.
1860 (w - 1)->wb_next = whead;
1862 /* Wait on any objects that aren't yet signalled. */
1864 /* Calculate timeout, if any. */
1866 if (duetime != NULL) {
1868 tv.tv_sec = - (*duetime) / 10000000;
1869 tv.tv_usec = (- (*duetime) / 10) -
1870 (tv.tv_sec * 1000000);
1872 ntoskrnl_time(&curtime);
1873 if (*duetime < curtime)
1874 tv.tv_sec = tv.tv_usec = 0;
1876 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1877 tv.tv_usec = ((*duetime) - curtime) / 10 -
1878 (tv.tv_sec * 1000000);
1886 if (duetime == NULL)
1887 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1889 error = cv_timedwait(&we.we_cv,
1890 &ntoskrnl_dispatchlock, tvtohz(&tv));
1892 /* Wait with timeout expired. */
1895 status = STATUS_TIMEOUT;
1901 /* See what's been signalled. */
1906 if (ntoskrnl_is_signalled(cur, td) == TRUE ||
1907 w->wb_awakened == TRUE) {
1908 /* Sanity check the signal state value. */
1909 if (cur->dh_sigstate == INT32_MIN &&
1910 cur->dh_type == DISP_TYPE_MUTANT) {
1911 mtx_unlock(&ntoskrnl_dispatchlock);
1912 panic("mutant limit exceeded");
1915 if (wtype == WAITTYPE_ANY) {
1916 status = w->wb_waitkey &
1922 } while (w != whead);
1925 * If all objects have been signalled, or if this
1926 * is a WAITTYPE_ANY wait and we were woke up by
1927 * someone, we can bail.
1931 status = STATUS_SUCCESS;
1936 * If this is WAITTYPE_ALL wait, and there's still
1937 * objects that haven't been signalled, deduct the
1938 * time that's elapsed so far from the timeout and
1939 * wait again (or continue waiting indefinitely if
1940 * there's no timeout).
1943 if (duetime != NULL) {
1944 tv.tv_sec -= (t2.tv_sec - t1.tv_sec);
1945 tv.tv_usec -= (t2.tv_nsec - t1.tv_nsec) / 1000;
1951 cv_destroy(&we.we_cv);
1953 for (i = 0; i < cnt; i++) {
1954 if (whead[i].wb_object != NULL)
1955 RemoveEntryList(&whead[i].wb_waitlist);
1957 mtx_unlock(&ntoskrnl_dispatchlock);
1963 WRITE_REGISTER_USHORT(uint16_t *reg, uint16_t val)
1965 bus_space_write_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1969 READ_REGISTER_USHORT(reg)
1972 return (bus_space_read_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1976 WRITE_REGISTER_ULONG(reg, val)
1980 bus_space_write_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1984 READ_REGISTER_ULONG(reg)
1987 return (bus_space_read_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1991 READ_REGISTER_UCHAR(uint8_t *reg)
1993 return (bus_space_read_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1997 WRITE_REGISTER_UCHAR(uint8_t *reg, uint8_t val)
1999 bus_space_write_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
2051 _allshl(int64_t a, uint8_t b)
2057 _aullshl(uint64_t a, uint8_t b)
2063 _allshr(int64_t a, uint8_t b)
2069 _aullshr(uint64_t a, uint8_t b)
2074 static slist_entry *
2075 ntoskrnl_pushsl(head, entry)
2079 slist_entry *oldhead;
2081 oldhead = head->slh_list.slh_next;
2082 entry->sl_next = head->slh_list.slh_next;
2083 head->slh_list.slh_next = entry;
2084 head->slh_list.slh_depth++;
2085 head->slh_list.slh_seq++;
2091 InitializeSListHead(head)
2094 memset(head, 0, sizeof(*head));
2097 static slist_entry *
2098 ntoskrnl_popsl(head)
2103 first = head->slh_list.slh_next;
2104 if (first != NULL) {
2105 head->slh_list.slh_next = first->sl_next;
2106 head->slh_list.slh_depth--;
2107 head->slh_list.slh_seq++;
2114 * We need this to make lookaside lists work for amd64.
2115 * We pass a pointer to ExAllocatePoolWithTag() the lookaside
2116 * list structure. For amd64 to work right, this has to be a
2117 * pointer to the wrapped version of the routine, not the
2118 * original. Letting the Windows driver invoke the original
2119 * function directly will result in a convention calling
2120 * mismatch and a pretty crash. On x86, this effectively
2121 * becomes a no-op since ipt_func and ipt_wrap are the same.
2125 ntoskrnl_findwrap(func)
2128 image_patch_table *patch;
2130 patch = ntoskrnl_functbl;
2131 while (patch->ipt_func != NULL) {
2132 if ((funcptr)patch->ipt_func == func)
2133 return ((funcptr)patch->ipt_wrap);
2141 ExInitializePagedLookasideList(paged_lookaside_list *lookaside,
2142 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2143 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2145 bzero((char *)lookaside, sizeof(paged_lookaside_list));
2147 if (size < sizeof(slist_entry))
2148 lookaside->nll_l.gl_size = sizeof(slist_entry);
2150 lookaside->nll_l.gl_size = size;
2151 lookaside->nll_l.gl_tag = tag;
2152 if (allocfunc == NULL)
2153 lookaside->nll_l.gl_allocfunc =
2154 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2156 lookaside->nll_l.gl_allocfunc = allocfunc;
2158 if (freefunc == NULL)
2159 lookaside->nll_l.gl_freefunc =
2160 ntoskrnl_findwrap((funcptr)ExFreePool);
2162 lookaside->nll_l.gl_freefunc = freefunc;
2165 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2168 lookaside->nll_l.gl_type = NonPagedPool;
2169 lookaside->nll_l.gl_depth = depth;
2170 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2174 ExDeletePagedLookasideList(lookaside)
2175 paged_lookaside_list *lookaside;
2178 void (*freefunc)(void *);
2180 freefunc = lookaside->nll_l.gl_freefunc;
2181 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2182 MSCALL1(freefunc, buf);
2186 ExInitializeNPagedLookasideList(npaged_lookaside_list *lookaside,
2187 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2188 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2190 bzero((char *)lookaside, sizeof(npaged_lookaside_list));
2192 if (size < sizeof(slist_entry))
2193 lookaside->nll_l.gl_size = sizeof(slist_entry);
2195 lookaside->nll_l.gl_size = size;
2196 lookaside->nll_l.gl_tag = tag;
2197 if (allocfunc == NULL)
2198 lookaside->nll_l.gl_allocfunc =
2199 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2201 lookaside->nll_l.gl_allocfunc = allocfunc;
2203 if (freefunc == NULL)
2204 lookaside->nll_l.gl_freefunc =
2205 ntoskrnl_findwrap((funcptr)ExFreePool);
2207 lookaside->nll_l.gl_freefunc = freefunc;
2210 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2213 lookaside->nll_l.gl_type = NonPagedPool;
2214 lookaside->nll_l.gl_depth = depth;
2215 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2219 ExDeleteNPagedLookasideList(lookaside)
2220 npaged_lookaside_list *lookaside;
2223 void (*freefunc)(void *);
2225 freefunc = lookaside->nll_l.gl_freefunc;
2226 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2227 MSCALL1(freefunc, buf);
2231 InterlockedPushEntrySList(head, entry)
2235 slist_entry *oldhead;
2237 mtx_lock_spin(&ntoskrnl_interlock);
2238 oldhead = ntoskrnl_pushsl(head, entry);
2239 mtx_unlock_spin(&ntoskrnl_interlock);
2245 InterlockedPopEntrySList(head)
2250 mtx_lock_spin(&ntoskrnl_interlock);
2251 first = ntoskrnl_popsl(head);
2252 mtx_unlock_spin(&ntoskrnl_interlock);
2257 static slist_entry *
2258 ExInterlockedPushEntrySList(head, entry, lock)
2263 return (InterlockedPushEntrySList(head, entry));
2266 static slist_entry *
2267 ExInterlockedPopEntrySList(head, lock)
2271 return (InterlockedPopEntrySList(head));
2275 ExQueryDepthSList(head)
2280 mtx_lock_spin(&ntoskrnl_interlock);
2281 depth = head->slh_list.slh_depth;
2282 mtx_unlock_spin(&ntoskrnl_interlock);
2288 KeInitializeSpinLock(lock)
2296 KefAcquireSpinLockAtDpcLevel(lock)
2299 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2303 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0) {
2305 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2314 KefReleaseSpinLockFromDpcLevel(lock)
2317 atomic_store_rel_int((volatile u_int *)lock, 0);
2321 KeAcquireSpinLockRaiseToDpc(kspin_lock *lock)
2325 if (KeGetCurrentIrql() > DISPATCH_LEVEL)
2326 panic("IRQL_NOT_LESS_THAN_OR_EQUAL");
2328 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
2329 KeAcquireSpinLockAtDpcLevel(lock);
2335 KeAcquireSpinLockAtDpcLevel(kspin_lock *lock)
2337 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0)
2342 KeReleaseSpinLockFromDpcLevel(kspin_lock *lock)
2344 atomic_store_rel_int((volatile u_int *)lock, 0);
2346 #endif /* __i386__ */
2349 InterlockedExchange(dst, val)
2350 volatile uint32_t *dst;
2355 mtx_lock_spin(&ntoskrnl_interlock);
2358 mtx_unlock_spin(&ntoskrnl_interlock);
2364 InterlockedIncrement(addend)
2365 volatile uint32_t *addend;
2367 atomic_add_long((volatile u_long *)addend, 1);
2372 InterlockedDecrement(addend)
2373 volatile uint32_t *addend;
2375 atomic_subtract_long((volatile u_long *)addend, 1);
2380 ExInterlockedAddLargeStatistic(addend, inc)
2384 mtx_lock_spin(&ntoskrnl_interlock);
2386 mtx_unlock_spin(&ntoskrnl_interlock);
2390 IoAllocateMdl(void *vaddr, uint32_t len, uint8_t secondarybuf,
2391 uint8_t chargequota, irp *iopkt)
2396 if (MmSizeOfMdl(vaddr, len) > MDL_ZONE_SIZE)
2397 m = ExAllocatePoolWithTag(NonPagedPool,
2398 MmSizeOfMdl(vaddr, len), 0);
2400 m = uma_zalloc(mdl_zone, M_NOWAIT | M_ZERO);
2407 MmInitializeMdl(m, vaddr, len);
2410 * MmInitializMdl() clears the flags field, so we
2411 * have to set this here. If the MDL came from the
2412 * MDL UMA zone, tag it so we can release it to
2413 * the right place later.
2416 m->mdl_flags = MDL_ZONE_ALLOCED;
2418 if (iopkt != NULL) {
2419 if (secondarybuf == TRUE) {
2421 last = iopkt->irp_mdl;
2422 while (last->mdl_next != NULL)
2423 last = last->mdl_next;
2426 if (iopkt->irp_mdl != NULL)
2427 panic("leaking an MDL in IoAllocateMdl()");
2442 if (m->mdl_flags & MDL_ZONE_ALLOCED)
2443 uma_zfree(mdl_zone, m);
2449 MmAllocateContiguousMemory(size, highest)
2454 size_t pagelength = roundup(size, PAGE_SIZE);
2456 addr = ExAllocatePoolWithTag(NonPagedPool, pagelength, 0);
2462 MmAllocateContiguousMemorySpecifyCache(size, lowest, highest,
2463 boundary, cachetype)
2468 enum nt_caching_type cachetype;
2470 vm_memattr_t memattr;
2473 switch (cachetype) {
2475 memattr = VM_MEMATTR_UNCACHEABLE;
2477 case MmWriteCombined:
2478 memattr = VM_MEMATTR_WRITE_COMBINING;
2480 case MmNonCachedUnordered:
2481 memattr = VM_MEMATTR_UNCACHEABLE;
2484 case MmHardwareCoherentCached:
2487 memattr = VM_MEMATTR_DEFAULT;
2491 ret = (void *)kmem_alloc_contig(size, M_ZERO | M_NOWAIT, lowest,
2492 highest, PAGE_SIZE, boundary, memattr);
2494 malloc_type_allocated(M_DEVBUF, round_page(size));
2499 MmFreeContiguousMemory(base)
2506 MmFreeContiguousMemorySpecifyCache(base, size, cachetype)
2509 enum nt_caching_type cachetype;
2511 contigfree(base, size, M_DEVBUF);
2515 MmSizeOfMdl(vaddr, len)
2521 l = sizeof(struct mdl) +
2522 (sizeof(vm_offset_t *) * SPAN_PAGES(vaddr, len));
2528 * The Microsoft documentation says this routine fills in the
2529 * page array of an MDL with the _physical_ page addresses that
2530 * comprise the buffer, but we don't really want to do that here.
2531 * Instead, we just fill in the page array with the kernel virtual
2532 * addresses of the buffers.
2535 MmBuildMdlForNonPagedPool(m)
2538 vm_offset_t *mdl_pages;
2541 pagecnt = SPAN_PAGES(m->mdl_byteoffset, m->mdl_bytecount);
2543 if (pagecnt > (m->mdl_size - sizeof(mdl)) / sizeof(vm_offset_t *))
2544 panic("not enough pages in MDL to describe buffer");
2546 mdl_pages = MmGetMdlPfnArray(m);
2548 for (i = 0; i < pagecnt; i++)
2549 *mdl_pages = (vm_offset_t)m->mdl_startva + (i * PAGE_SIZE);
2551 m->mdl_flags |= MDL_SOURCE_IS_NONPAGED_POOL;
2552 m->mdl_mappedsystemva = MmGetMdlVirtualAddress(m);
2556 MmMapLockedPages(mdl *buf, uint8_t accessmode)
2558 buf->mdl_flags |= MDL_MAPPED_TO_SYSTEM_VA;
2559 return (MmGetMdlVirtualAddress(buf));
2563 MmMapLockedPagesSpecifyCache(mdl *buf, uint8_t accessmode, uint32_t cachetype,
2564 void *vaddr, uint32_t bugcheck, uint32_t prio)
2566 return (MmMapLockedPages(buf, accessmode));
2570 MmUnmapLockedPages(vaddr, buf)
2574 buf->mdl_flags &= ~MDL_MAPPED_TO_SYSTEM_VA;
2578 * This function has a problem in that it will break if you
2579 * compile this module without PAE and try to use it on a PAE
2580 * kernel. Unfortunately, there's no way around this at the
2581 * moment. It's slightly less broken that using pmap_kextract().
2582 * You'd think the virtual memory subsystem would help us out
2583 * here, but it doesn't.
2587 MmGetPhysicalAddress(void *base)
2589 return (pmap_extract(kernel_map->pmap, (vm_offset_t)base));
2593 MmGetSystemRoutineAddress(ustr)
2594 unicode_string *ustr;
2598 if (RtlUnicodeStringToAnsiString(&astr, ustr, TRUE))
2600 return (ndis_get_routine_address(ntoskrnl_functbl, astr.as_buf));
2604 MmIsAddressValid(vaddr)
2607 if (pmap_extract(kernel_map->pmap, (vm_offset_t)vaddr))
2614 MmMapIoSpace(paddr, len, cachetype)
2619 devclass_t nexus_class;
2620 device_t *nexus_devs, devp;
2621 int nexus_count = 0;
2622 device_t matching_dev = NULL;
2623 struct resource *res;
2627 /* There will always be at least one nexus. */
2629 nexus_class = devclass_find("nexus");
2630 devclass_get_devices(nexus_class, &nexus_devs, &nexus_count);
2632 for (i = 0; i < nexus_count; i++) {
2633 devp = nexus_devs[i];
2634 matching_dev = ntoskrnl_finddev(devp, paddr, &res);
2639 free(nexus_devs, M_TEMP);
2641 if (matching_dev == NULL)
2644 v = (vm_offset_t)rman_get_virtual(res);
2645 if (paddr > rman_get_start(res))
2646 v += paddr - rman_get_start(res);
2652 MmUnmapIoSpace(vaddr, len)
2659 ntoskrnl_finddev(dev, paddr, res)
2662 struct resource **res;
2664 device_t *children = NULL;
2665 device_t matching_dev;
2668 struct resource_list *rl;
2669 struct resource_list_entry *rle;
2673 /* We only want devices that have been successfully probed. */
2675 if (device_is_alive(dev) == FALSE)
2678 rl = BUS_GET_RESOURCE_LIST(device_get_parent(dev), dev);
2680 STAILQ_FOREACH(rle, rl, link) {
2686 flags = rman_get_flags(r);
2688 if (rle->type == SYS_RES_MEMORY &&
2689 paddr >= rman_get_start(r) &&
2690 paddr <= rman_get_end(r)) {
2691 if (!(flags & RF_ACTIVE))
2692 bus_activate_resource(dev,
2693 SYS_RES_MEMORY, 0, r);
2701 * If this device has children, do another
2702 * level of recursion to inspect them.
2705 device_get_children(dev, &children, &childcnt);
2707 for (i = 0; i < childcnt; i++) {
2708 matching_dev = ntoskrnl_finddev(children[i], paddr, res);
2709 if (matching_dev != NULL) {
2710 free(children, M_TEMP);
2711 return (matching_dev);
2715 /* Won't somebody please think of the children! */
2717 if (children != NULL)
2718 free(children, M_TEMP);
2724 * Workitems are unlike DPCs, in that they run in a user-mode thread
2725 * context rather than at DISPATCH_LEVEL in kernel context. In our
2726 * case we run them in kernel context anyway.
2729 ntoskrnl_workitem_thread(arg)
2739 InitializeListHead(&kq->kq_disp);
2740 kq->kq_td = curthread;
2742 KeInitializeSpinLock(&kq->kq_lock);
2743 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
2746 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
2748 KeAcquireSpinLock(&kq->kq_lock, &irql);
2752 KeReleaseSpinLock(&kq->kq_lock, irql);
2756 while (!IsListEmpty(&kq->kq_disp)) {
2757 l = RemoveHeadList(&kq->kq_disp);
2758 iw = CONTAINING_RECORD(l,
2759 io_workitem, iw_listentry);
2760 InitializeListHead((&iw->iw_listentry));
2761 if (iw->iw_func == NULL)
2763 KeReleaseSpinLock(&kq->kq_lock, irql);
2764 MSCALL2(iw->iw_func, iw->iw_dobj, iw->iw_ctx);
2765 KeAcquireSpinLock(&kq->kq_lock, &irql);
2768 KeReleaseSpinLock(&kq->kq_lock, irql);
2772 return; /* notreached */
2776 RtlCharToInteger(src, base, val)
2785 return (STATUS_ACCESS_VIOLATION);
2786 while (*src != '\0' && *src <= ' ')
2790 else if (*src == '-') {
2801 } else if (*src == 'o') {
2804 } else if (*src == 'x') {
2809 } else if (!(base == 2 || base == 8 || base == 10 || base == 16))
2810 return (STATUS_INVALID_PARAMETER);
2812 for (res = 0; *src; src++) {
2816 else if (isxdigit(*src))
2817 v = tolower(*src) - 'a' + 10;
2821 return (STATUS_INVALID_PARAMETER);
2822 res = res * base + v;
2824 *val = negative ? -res : res;
2825 return (STATUS_SUCCESS);
2829 ntoskrnl_destroy_workitem_threads(void)
2834 for (i = 0; i < WORKITEM_THREADS; i++) {
2837 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2839 tsleep(kq->kq_td->td_proc, PWAIT, "waitiw", hz/10);
2844 IoAllocateWorkItem(dobj)
2845 device_object *dobj;
2849 iw = uma_zalloc(iw_zone, M_NOWAIT);
2853 InitializeListHead(&iw->iw_listentry);
2856 mtx_lock(&ntoskrnl_dispatchlock);
2857 iw->iw_idx = wq_idx;
2858 WORKIDX_INC(wq_idx);
2859 mtx_unlock(&ntoskrnl_dispatchlock);
2868 uma_zfree(iw_zone, iw);
2872 IoQueueWorkItem(iw, iw_func, qtype, ctx)
2874 io_workitem_func iw_func;
2883 kq = wq_queues + iw->iw_idx;
2885 KeAcquireSpinLock(&kq->kq_lock, &irql);
2888 * Traverse the list and make sure this workitem hasn't
2889 * already been inserted. Queuing the same workitem
2890 * twice will hose the list but good.
2893 l = kq->kq_disp.nle_flink;
2894 while (l != &kq->kq_disp) {
2895 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2897 /* Already queued -- do nothing. */
2898 KeReleaseSpinLock(&kq->kq_lock, irql);
2904 iw->iw_func = iw_func;
2907 InsertTailList((&kq->kq_disp), (&iw->iw_listentry));
2908 KeReleaseSpinLock(&kq->kq_lock, irql);
2910 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2914 ntoskrnl_workitem(dobj, arg)
2915 device_object *dobj;
2923 w = (work_queue_item *)dobj;
2924 f = (work_item_func)w->wqi_func;
2925 uma_zfree(iw_zone, iw);
2926 MSCALL2(f, w, w->wqi_ctx);
2930 * The ExQueueWorkItem() API is deprecated in Windows XP. Microsoft
2931 * warns that it's unsafe and to use IoQueueWorkItem() instead. The
2932 * problem with ExQueueWorkItem() is that it can't guard against
2933 * the condition where a driver submits a job to the work queue and
2934 * is then unloaded before the job is able to run. IoQueueWorkItem()
2935 * acquires a reference to the device's device_object via the
2936 * object manager and retains it until after the job has completed,
2937 * which prevents the driver from being unloaded before the job
2938 * runs. (We don't currently support this behavior, though hopefully
2939 * that will change once the object manager API is fleshed out a bit.)
2941 * Having said all that, the ExQueueWorkItem() API remains, because
2942 * there are still other parts of Windows that use it, including
2943 * NDIS itself: NdisScheduleWorkItem() calls ExQueueWorkItem().
2944 * We fake up the ExQueueWorkItem() API on top of our implementation
2945 * of IoQueueWorkItem(). Workitem thread #3 is reserved exclusively
2946 * for ExQueueWorkItem() jobs, and we pass a pointer to the work
2947 * queue item (provided by the caller) in to IoAllocateWorkItem()
2948 * instead of the device_object. We need to save this pointer so
2949 * we can apply a sanity check: as with the DPC queue and other
2950 * workitem queues, we can't allow the same work queue item to
2951 * be queued twice. If it's already pending, we silently return
2955 ExQueueWorkItem(w, qtype)
2960 io_workitem_func iwf;
2967 * We need to do a special sanity test to make sure
2968 * the ExQueueWorkItem() API isn't used to queue
2969 * the same workitem twice. Rather than checking the
2970 * io_workitem pointer itself, we test the attached
2971 * device object, which is really a pointer to the
2972 * legacy work queue item structure.
2975 kq = wq_queues + WORKITEM_LEGACY_THREAD;
2976 KeAcquireSpinLock(&kq->kq_lock, &irql);
2977 l = kq->kq_disp.nle_flink;
2978 while (l != &kq->kq_disp) {
2979 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2980 if (cur->iw_dobj == (device_object *)w) {
2981 /* Already queued -- do nothing. */
2982 KeReleaseSpinLock(&kq->kq_lock, irql);
2987 KeReleaseSpinLock(&kq->kq_lock, irql);
2989 iw = IoAllocateWorkItem((device_object *)w);
2993 iw->iw_idx = WORKITEM_LEGACY_THREAD;
2994 iwf = (io_workitem_func)ntoskrnl_findwrap((funcptr)ntoskrnl_workitem);
2995 IoQueueWorkItem(iw, iwf, qtype, iw);
2999 RtlZeroMemory(dst, len)
3007 RtlSecureZeroMemory(dst, len)
3011 memset(dst, 0, len);
3015 RtlFillMemory(void *dst, size_t len, uint8_t c)
3017 memset(dst, c, len);
3021 RtlMoveMemory(dst, src, len)
3026 memmove(dst, src, len);
3030 RtlCopyMemory(dst, src, len)
3035 bcopy(src, dst, len);
3039 RtlCompareMemory(s1, s2, len)
3047 m1 = __DECONST(char *, s1);
3048 m2 = __DECONST(char *, s2);
3050 for (i = 0; i < len && m1[i] == m2[i]; i++);
3055 RtlInitAnsiString(dst, src)
3065 a->as_len = a->as_maxlen = 0;
3069 a->as_len = a->as_maxlen = strlen(src);
3074 RtlInitUnicodeString(dst, src)
3075 unicode_string *dst;
3085 u->us_len = u->us_maxlen = 0;
3092 u->us_len = u->us_maxlen = i * 2;
3097 RtlUnicodeStringToInteger(ustr, base, val)
3098 unicode_string *ustr;
3107 uchr = ustr->us_buf;
3109 bzero(abuf, sizeof(abuf));
3111 if ((char)((*uchr) & 0xFF) == '-') {
3115 } else if ((char)((*uchr) & 0xFF) == '+') {
3122 if ((char)((*uchr) & 0xFF) == 'b') {
3126 } else if ((char)((*uchr) & 0xFF) == 'o') {
3130 } else if ((char)((*uchr) & 0xFF) == 'x') {
3144 ntoskrnl_unicode_to_ascii(uchr, astr, len);
3145 *val = strtoul(abuf, NULL, base);
3147 return (STATUS_SUCCESS);
3151 RtlFreeUnicodeString(ustr)
3152 unicode_string *ustr;
3154 if (ustr->us_buf == NULL)
3156 ExFreePool(ustr->us_buf);
3157 ustr->us_buf = NULL;
3161 RtlFreeAnsiString(astr)
3164 if (astr->as_buf == NULL)
3166 ExFreePool(astr->as_buf);
3167 astr->as_buf = NULL;
3174 return (int)strtol(str, (char **)NULL, 10);
3181 return strtol(str, (char **)NULL, 10);
3192 srand(unsigned int seed __unused)
3197 IoIsWdmVersionAvailable(uint8_t major, uint8_t minor)
3199 if (major == WDM_MAJOR && minor == WDM_MINOR_WINXP)
3205 IoOpenDeviceRegistryKey(struct device_object *devobj, uint32_t type,
3206 uint32_t mask, void **key)
3208 return (NDIS_STATUS_INVALID_DEVICE_REQUEST);
3212 IoGetDeviceObjectPointer(name, reqaccess, fileobj, devobj)
3213 unicode_string *name;
3216 device_object *devobj;
3218 return (STATUS_SUCCESS);
3222 IoGetDeviceProperty(devobj, regprop, buflen, prop, reslen)
3223 device_object *devobj;
3232 drv = devobj->do_drvobj;
3235 case DEVPROP_DRIVER_KEYNAME:
3237 *name = drv->dro_drivername.us_buf;
3238 *reslen = drv->dro_drivername.us_len;
3241 return (STATUS_INVALID_PARAMETER_2);
3245 return (STATUS_SUCCESS);
3249 KeInitializeMutex(kmutex, level)
3253 InitializeListHead((&kmutex->km_header.dh_waitlisthead));
3254 kmutex->km_abandoned = FALSE;
3255 kmutex->km_apcdisable = 1;
3256 kmutex->km_header.dh_sigstate = 1;
3257 kmutex->km_header.dh_type = DISP_TYPE_MUTANT;
3258 kmutex->km_header.dh_size = sizeof(kmutant) / sizeof(uint32_t);
3259 kmutex->km_ownerthread = NULL;
3263 KeReleaseMutex(kmutant *kmutex, uint8_t kwait)
3267 mtx_lock(&ntoskrnl_dispatchlock);
3268 prevstate = kmutex->km_header.dh_sigstate;
3269 if (kmutex->km_ownerthread != curthread) {
3270 mtx_unlock(&ntoskrnl_dispatchlock);
3271 return (STATUS_MUTANT_NOT_OWNED);
3274 kmutex->km_header.dh_sigstate++;
3275 kmutex->km_abandoned = FALSE;
3277 if (kmutex->km_header.dh_sigstate == 1) {
3278 kmutex->km_ownerthread = NULL;
3279 ntoskrnl_waittest(&kmutex->km_header, IO_NO_INCREMENT);
3282 mtx_unlock(&ntoskrnl_dispatchlock);
3288 KeReadStateMutex(kmutex)
3291 return (kmutex->km_header.dh_sigstate);
3295 KeInitializeEvent(nt_kevent *kevent, uint32_t type, uint8_t state)
3297 InitializeListHead((&kevent->k_header.dh_waitlisthead));
3298 kevent->k_header.dh_sigstate = state;
3299 if (type == EVENT_TYPE_NOTIFY)
3300 kevent->k_header.dh_type = DISP_TYPE_NOTIFICATION_EVENT;
3302 kevent->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_EVENT;
3303 kevent->k_header.dh_size = sizeof(nt_kevent) / sizeof(uint32_t);
3307 KeResetEvent(kevent)
3312 mtx_lock(&ntoskrnl_dispatchlock);
3313 prevstate = kevent->k_header.dh_sigstate;
3314 kevent->k_header.dh_sigstate = FALSE;
3315 mtx_unlock(&ntoskrnl_dispatchlock);
3321 KeSetEvent(nt_kevent *kevent, uint32_t increment, uint8_t kwait)
3325 nt_dispatch_header *dh;
3329 mtx_lock(&ntoskrnl_dispatchlock);
3330 prevstate = kevent->k_header.dh_sigstate;
3331 dh = &kevent->k_header;
3333 if (IsListEmpty(&dh->dh_waitlisthead))
3335 * If there's nobody in the waitlist, just set
3336 * the state to signalled.
3338 dh->dh_sigstate = 1;
3341 * Get the first waiter. If this is a synchronization
3342 * event, just wake up that one thread (don't bother
3343 * setting the state to signalled since we're supposed
3344 * to automatically clear synchronization events anyway).
3346 * If it's a notification event, or the first
3347 * waiter is doing a WAITTYPE_ALL wait, go through
3348 * the full wait satisfaction process.
3350 w = CONTAINING_RECORD(dh->dh_waitlisthead.nle_flink,
3351 wait_block, wb_waitlist);
3354 if (kevent->k_header.dh_type == DISP_TYPE_NOTIFICATION_EVENT ||
3355 w->wb_waittype == WAITTYPE_ALL) {
3356 if (prevstate == 0) {
3357 dh->dh_sigstate = 1;
3358 ntoskrnl_waittest(dh, increment);
3361 w->wb_awakened |= TRUE;
3362 cv_broadcastpri(&we->we_cv,
3363 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
3364 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
3368 mtx_unlock(&ntoskrnl_dispatchlock);
3374 KeClearEvent(kevent)
3377 kevent->k_header.dh_sigstate = FALSE;
3381 KeReadStateEvent(kevent)
3384 return (kevent->k_header.dh_sigstate);
3388 * The object manager in Windows is responsible for managing
3389 * references and access to various types of objects, including
3390 * device_objects, events, threads, timers and so on. However,
3391 * there's a difference in the way objects are handled in user
3392 * mode versus kernel mode.
3394 * In user mode (i.e. Win32 applications), all objects are
3395 * managed by the object manager. For example, when you create
3396 * a timer or event object, you actually end up with an
3397 * object_header (for the object manager's bookkeeping
3398 * purposes) and an object body (which contains the actual object
3399 * structure, e.g. ktimer, kevent, etc...). This allows Windows
3400 * to manage resource quotas and to enforce access restrictions
3401 * on basically every kind of system object handled by the kernel.
3403 * However, in kernel mode, you only end up using the object
3404 * manager some of the time. For example, in a driver, you create
3405 * a timer object by simply allocating the memory for a ktimer
3406 * structure and initializing it with KeInitializeTimer(). Hence,
3407 * the timer has no object_header and no reference counting or
3408 * security/resource checks are done on it. The assumption in
3409 * this case is that if you're running in kernel mode, you know
3410 * what you're doing, and you're already at an elevated privilege
3413 * There are some exceptions to this. The two most important ones
3414 * for our purposes are device_objects and threads. We need to use
3415 * the object manager to do reference counting on device_objects,
3416 * and for threads, you can only get a pointer to a thread's
3417 * dispatch header by using ObReferenceObjectByHandle() on the
3418 * handle returned by PsCreateSystemThread().
3422 ObReferenceObjectByHandle(ndis_handle handle, uint32_t reqaccess, void *otype,
3423 uint8_t accessmode, void **object, void **handleinfo)
3427 nr = malloc(sizeof(nt_objref), M_DEVBUF, M_NOWAIT|M_ZERO);
3429 return (STATUS_INSUFFICIENT_RESOURCES);
3431 InitializeListHead((&nr->no_dh.dh_waitlisthead));
3432 nr->no_obj = handle;
3433 nr->no_dh.dh_type = DISP_TYPE_THREAD;
3434 nr->no_dh.dh_sigstate = 0;
3435 nr->no_dh.dh_size = (uint8_t)(sizeof(struct thread) /
3437 TAILQ_INSERT_TAIL(&ntoskrnl_reflist, nr, link);
3440 return (STATUS_SUCCESS);
3444 ObfDereferenceObject(object)
3450 TAILQ_REMOVE(&ntoskrnl_reflist, nr, link);
3458 return (STATUS_SUCCESS);
3462 WmiQueryTraceInformation(traceclass, traceinfo, infolen, reqlen, buf)
3463 uint32_t traceclass;
3469 return (STATUS_NOT_FOUND);
3473 WmiTraceMessage(uint64_t loghandle, uint32_t messageflags,
3474 void *guid, uint16_t messagenum, ...)
3476 return (STATUS_SUCCESS);
3480 IoWMIRegistrationControl(dobj, action)
3481 device_object *dobj;
3484 return (STATUS_SUCCESS);
3488 * This is here just in case the thread returns without calling
3489 * PsTerminateSystemThread().
3492 ntoskrnl_thrfunc(arg)
3495 thread_context *thrctx;
3496 uint32_t (*tfunc)(void *);
3501 tfunc = thrctx->tc_thrfunc;
3502 tctx = thrctx->tc_thrctx;
3503 free(thrctx, M_TEMP);
3505 rval = MSCALL1(tfunc, tctx);
3507 PsTerminateSystemThread(rval);
3508 return; /* notreached */
3512 PsCreateSystemThread(handle, reqaccess, objattrs, phandle,
3513 clientid, thrfunc, thrctx)
3514 ndis_handle *handle;
3517 ndis_handle phandle;
3526 tc = malloc(sizeof(thread_context), M_TEMP, M_NOWAIT);
3528 return (STATUS_INSUFFICIENT_RESOURCES);
3530 tc->tc_thrctx = thrctx;
3531 tc->tc_thrfunc = thrfunc;
3533 error = kproc_create(ntoskrnl_thrfunc, tc, &p,
3534 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Kthread %d", ntoskrnl_kth);
3538 return (STATUS_INSUFFICIENT_RESOURCES);
3544 return (STATUS_SUCCESS);
3548 * In Windows, the exit of a thread is an event that you're allowed
3549 * to wait on, assuming you've obtained a reference to the thread using
3550 * ObReferenceObjectByHandle(). Unfortunately, the only way we can
3551 * simulate this behavior is to register each thread we create in a
3552 * reference list, and if someone holds a reference to us, we poke
3556 PsTerminateSystemThread(status)
3559 struct nt_objref *nr;
3561 mtx_lock(&ntoskrnl_dispatchlock);
3562 TAILQ_FOREACH(nr, &ntoskrnl_reflist, link) {
3563 if (nr->no_obj != curthread->td_proc)
3565 nr->no_dh.dh_sigstate = 1;
3566 ntoskrnl_waittest(&nr->no_dh, IO_NO_INCREMENT);
3569 mtx_unlock(&ntoskrnl_dispatchlock);
3574 return (0); /* notreached */
3578 DbgPrint(char *fmt, ...)
3588 return (STATUS_SUCCESS);
3595 kdb_enter(KDB_WHY_NDIS, "DbgBreakPoint(): breakpoint");
3599 KeBugCheckEx(code, param1, param2, param3, param4)
3606 panic("KeBugCheckEx: STOP 0x%X", code);
3610 ntoskrnl_timercall(arg)
3617 mtx_lock(&ntoskrnl_dispatchlock);
3621 #ifdef NTOSKRNL_DEBUG_TIMERS
3622 ntoskrnl_timer_fires++;
3624 ntoskrnl_remove_timer(timer);
3627 * This should never happen, but complain
3631 if (timer->k_header.dh_inserted == FALSE) {
3632 mtx_unlock(&ntoskrnl_dispatchlock);
3633 printf("NTOS: timer %p fired even though "
3634 "it was canceled\n", timer);
3638 /* Mark the timer as no longer being on the timer queue. */
3640 timer->k_header.dh_inserted = FALSE;
3642 /* Now signal the object and satisfy any waits on it. */
3644 timer->k_header.dh_sigstate = 1;
3645 ntoskrnl_waittest(&timer->k_header, IO_NO_INCREMENT);
3648 * If this is a periodic timer, re-arm it
3649 * so it will fire again. We do this before
3650 * calling any deferred procedure calls because
3651 * it's possible the DPC might cancel the timer,
3652 * in which case it would be wrong for us to
3653 * re-arm it again afterwards.
3656 if (timer->k_period) {
3658 tv.tv_usec = timer->k_period * 1000;
3659 timer->k_header.dh_inserted = TRUE;
3660 ntoskrnl_insert_timer(timer, tvtohz(&tv));
3661 #ifdef NTOSKRNL_DEBUG_TIMERS
3662 ntoskrnl_timer_reloads++;
3668 mtx_unlock(&ntoskrnl_dispatchlock);
3670 /* If there's a DPC associated with the timer, queue it up. */
3673 KeInsertQueueDpc(dpc, NULL, NULL);
3676 #ifdef NTOSKRNL_DEBUG_TIMERS
3678 sysctl_show_timers(SYSCTL_HANDLER_ARGS)
3683 ntoskrnl_show_timers();
3684 return (sysctl_handle_int(oidp, &ret, 0, req));
3688 ntoskrnl_show_timers()
3693 mtx_lock_spin(&ntoskrnl_calllock);
3694 l = ntoskrnl_calllist.nle_flink;
3695 while(l != &ntoskrnl_calllist) {
3699 mtx_unlock_spin(&ntoskrnl_calllock);
3702 printf("%d timers available (out of %d)\n", i, NTOSKRNL_TIMEOUTS);
3703 printf("timer sets: %qu\n", ntoskrnl_timer_sets);
3704 printf("timer reloads: %qu\n", ntoskrnl_timer_reloads);
3705 printf("timer cancels: %qu\n", ntoskrnl_timer_cancels);
3706 printf("timer fires: %qu\n", ntoskrnl_timer_fires);
3712 * Must be called with dispatcher lock held.
3716 ntoskrnl_insert_timer(timer, ticks)
3725 * Try and allocate a timer.
3727 mtx_lock_spin(&ntoskrnl_calllock);
3728 if (IsListEmpty(&ntoskrnl_calllist)) {
3729 mtx_unlock_spin(&ntoskrnl_calllock);
3730 #ifdef NTOSKRNL_DEBUG_TIMERS
3731 ntoskrnl_show_timers();
3733 panic("out of timers!");
3735 l = RemoveHeadList(&ntoskrnl_calllist);
3736 mtx_unlock_spin(&ntoskrnl_calllock);
3738 e = CONTAINING_RECORD(l, callout_entry, ce_list);
3741 timer->k_callout = c;
3744 callout_reset(c, ticks, ntoskrnl_timercall, timer);
3748 ntoskrnl_remove_timer(timer)
3753 e = (callout_entry *)timer->k_callout;
3754 callout_stop(timer->k_callout);
3756 mtx_lock_spin(&ntoskrnl_calllock);
3757 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
3758 mtx_unlock_spin(&ntoskrnl_calllock);
3762 KeInitializeTimer(timer)
3768 KeInitializeTimerEx(timer, EVENT_TYPE_NOTIFY);
3772 KeInitializeTimerEx(timer, type)
3779 bzero((char *)timer, sizeof(ktimer));
3780 InitializeListHead((&timer->k_header.dh_waitlisthead));
3781 timer->k_header.dh_sigstate = FALSE;
3782 timer->k_header.dh_inserted = FALSE;
3783 if (type == EVENT_TYPE_NOTIFY)
3784 timer->k_header.dh_type = DISP_TYPE_NOTIFICATION_TIMER;
3786 timer->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_TIMER;
3787 timer->k_header.dh_size = sizeof(ktimer) / sizeof(uint32_t);
3791 * DPC subsystem. A Windows Defered Procedure Call has the following
3793 * - It runs at DISPATCH_LEVEL.
3794 * - It can have one of 3 importance values that control when it
3795 * runs relative to other DPCs in the queue.
3796 * - On SMP systems, it can be set to run on a specific processor.
3797 * In order to satisfy the last property, we create a DPC thread for
3798 * each CPU in the system and bind it to that CPU. Each thread
3799 * maintains three queues with different importance levels, which
3800 * will be processed in order from lowest to highest.
3802 * In Windows, interrupt handlers run as DPCs. (Not to be confused
3803 * with ISRs, which run in interrupt context and can preempt DPCs.)
3804 * ISRs are given the highest importance so that they'll take
3805 * precedence over timers and other things.
3809 ntoskrnl_dpc_thread(arg)
3819 InitializeListHead(&kq->kq_disp);
3820 kq->kq_td = curthread;
3822 kq->kq_running = FALSE;
3823 KeInitializeSpinLock(&kq->kq_lock);
3824 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
3825 KeInitializeEvent(&kq->kq_done, EVENT_TYPE_SYNC, FALSE);
3828 * Elevate our priority. DPCs are used to run interrupt
3829 * handlers, and they should trigger as soon as possible
3830 * once scheduled by an ISR.
3833 thread_lock(curthread);
3834 #ifdef NTOSKRNL_MULTIPLE_DPCS
3835 sched_bind(curthread, kq->kq_cpu);
3837 sched_prio(curthread, PRI_MIN_KERN);
3838 thread_unlock(curthread);
3841 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
3843 KeAcquireSpinLock(&kq->kq_lock, &irql);
3847 KeReleaseSpinLock(&kq->kq_lock, irql);
3851 kq->kq_running = TRUE;
3853 while (!IsListEmpty(&kq->kq_disp)) {
3854 l = RemoveHeadList((&kq->kq_disp));
3855 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3856 InitializeListHead((&d->k_dpclistentry));
3857 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
3858 MSCALL4(d->k_deferedfunc, d, d->k_deferredctx,
3859 d->k_sysarg1, d->k_sysarg2);
3860 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3863 kq->kq_running = FALSE;
3865 KeReleaseSpinLock(&kq->kq_lock, irql);
3867 KeSetEvent(&kq->kq_done, IO_NO_INCREMENT, FALSE);
3871 return; /* notreached */
3875 ntoskrnl_destroy_dpc_threads(void)
3882 #ifdef NTOSKRNL_MULTIPLE_DPCS
3883 for (i = 0; i < mp_ncpus; i++) {
3885 for (i = 0; i < 1; i++) {
3890 KeInitializeDpc(&dpc, NULL, NULL);
3891 KeSetTargetProcessorDpc(&dpc, i);
3892 KeInsertQueueDpc(&dpc, NULL, NULL);
3894 tsleep(kq->kq_td->td_proc, PWAIT, "dpcw", hz/10);
3899 ntoskrnl_insert_dpc(head, dpc)
3906 l = head->nle_flink;
3908 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3914 if (dpc->k_importance == KDPC_IMPORTANCE_LOW)
3915 InsertTailList((head), (&dpc->k_dpclistentry));
3917 InsertHeadList((head), (&dpc->k_dpclistentry));
3923 KeInitializeDpc(dpc, dpcfunc, dpcctx)
3932 dpc->k_deferedfunc = dpcfunc;
3933 dpc->k_deferredctx = dpcctx;
3934 dpc->k_num = KDPC_CPU_DEFAULT;
3935 dpc->k_importance = KDPC_IMPORTANCE_MEDIUM;
3936 InitializeListHead((&dpc->k_dpclistentry));
3940 KeInsertQueueDpc(dpc, sysarg1, sysarg2)
3954 #ifdef NTOSKRNL_MULTIPLE_DPCS
3955 KeRaiseIrql(DISPATCH_LEVEL, &irql);
3958 * By default, the DPC is queued to run on the same CPU
3959 * that scheduled it.
3962 if (dpc->k_num == KDPC_CPU_DEFAULT)
3963 kq += curthread->td_oncpu;
3966 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3968 KeAcquireSpinLock(&kq->kq_lock, &irql);
3971 r = ntoskrnl_insert_dpc(&kq->kq_disp, dpc);
3973 dpc->k_sysarg1 = sysarg1;
3974 dpc->k_sysarg2 = sysarg2;
3976 KeReleaseSpinLock(&kq->kq_lock, irql);
3981 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
3987 KeRemoveQueueDpc(dpc)
3996 #ifdef NTOSKRNL_MULTIPLE_DPCS
3997 KeRaiseIrql(DISPATCH_LEVEL, &irql);
3999 kq = kq_queues + dpc->k_num;
4001 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
4004 KeAcquireSpinLock(&kq->kq_lock, &irql);
4007 if (dpc->k_dpclistentry.nle_flink == &dpc->k_dpclistentry) {
4008 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
4013 RemoveEntryList((&dpc->k_dpclistentry));
4014 InitializeListHead((&dpc->k_dpclistentry));
4016 KeReleaseSpinLock(&kq->kq_lock, irql);
4022 KeSetImportanceDpc(dpc, imp)
4026 if (imp != KDPC_IMPORTANCE_LOW &&
4027 imp != KDPC_IMPORTANCE_MEDIUM &&
4028 imp != KDPC_IMPORTANCE_HIGH)
4031 dpc->k_importance = (uint8_t)imp;
4035 KeSetTargetProcessorDpc(kdpc *dpc, uint8_t cpu)
4044 KeFlushQueuedDpcs(void)
4050 * Poke each DPC queue and wait
4051 * for them to drain.
4054 #ifdef NTOSKRNL_MULTIPLE_DPCS
4055 for (i = 0; i < mp_ncpus; i++) {
4057 for (i = 0; i < 1; i++) {
4060 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
4061 KeWaitForSingleObject(&kq->kq_done, 0, 0, TRUE, NULL);
4066 KeGetCurrentProcessorNumber(void)
4068 return ((uint32_t)curthread->td_oncpu);
4072 KeSetTimerEx(timer, duetime, period, dpc)
4085 mtx_lock(&ntoskrnl_dispatchlock);
4087 if (timer->k_header.dh_inserted == TRUE) {
4088 ntoskrnl_remove_timer(timer);
4089 #ifdef NTOSKRNL_DEBUG_TIMERS
4090 ntoskrnl_timer_cancels++;
4092 timer->k_header.dh_inserted = FALSE;
4097 timer->k_duetime = duetime;
4098 timer->k_period = period;
4099 timer->k_header.dh_sigstate = FALSE;
4103 tv.tv_sec = - (duetime) / 10000000;
4104 tv.tv_usec = (- (duetime) / 10) -
4105 (tv.tv_sec * 1000000);
4107 ntoskrnl_time(&curtime);
4108 if (duetime < curtime)
4109 tv.tv_sec = tv.tv_usec = 0;
4111 tv.tv_sec = ((duetime) - curtime) / 10000000;
4112 tv.tv_usec = ((duetime) - curtime) / 10 -
4113 (tv.tv_sec * 1000000);
4117 timer->k_header.dh_inserted = TRUE;
4118 ntoskrnl_insert_timer(timer, tvtohz(&tv));
4119 #ifdef NTOSKRNL_DEBUG_TIMERS
4120 ntoskrnl_timer_sets++;
4123 mtx_unlock(&ntoskrnl_dispatchlock);
4129 KeSetTimer(timer, duetime, dpc)
4134 return (KeSetTimerEx(timer, duetime, 0, dpc));
4138 * The Windows DDK documentation seems to say that cancelling
4139 * a timer that has a DPC will result in the DPC also being
4140 * cancelled, but this isn't really the case.
4144 KeCancelTimer(timer)
4152 mtx_lock(&ntoskrnl_dispatchlock);
4154 pending = timer->k_header.dh_inserted;
4156 if (timer->k_header.dh_inserted == TRUE) {
4157 timer->k_header.dh_inserted = FALSE;
4158 ntoskrnl_remove_timer(timer);
4159 #ifdef NTOSKRNL_DEBUG_TIMERS
4160 ntoskrnl_timer_cancels++;
4164 mtx_unlock(&ntoskrnl_dispatchlock);
4170 KeReadStateTimer(timer)
4173 return (timer->k_header.dh_sigstate);
4177 KeDelayExecutionThread(uint8_t wait_mode, uint8_t alertable, int64_t *interval)
4182 panic("invalid wait_mode %d", wait_mode);
4184 KeInitializeTimer(&timer);
4185 KeSetTimer(&timer, *interval, NULL);
4186 KeWaitForSingleObject(&timer, 0, 0, alertable, NULL);
4188 return STATUS_SUCCESS;
4192 KeQueryInterruptTime(void)
4197 getmicrouptime(&tv);
4199 ticks = tvtohz(&tv);
4201 return ticks * howmany(10000000, hz);
4204 static struct thread *
4205 KeGetCurrentThread(void)
4212 KeSetPriorityThread(td, pri)
4219 return LOW_REALTIME_PRIORITY;
4221 if (td->td_priority <= PRI_MIN_KERN)
4222 old = HIGH_PRIORITY;
4223 else if (td->td_priority >= PRI_MAX_KERN)
4226 old = LOW_REALTIME_PRIORITY;
4229 if (pri == HIGH_PRIORITY)
4230 sched_prio(td, PRI_MIN_KERN);
4231 if (pri == LOW_REALTIME_PRIORITY)
4232 sched_prio(td, PRI_MIN_KERN + (PRI_MAX_KERN - PRI_MIN_KERN) / 2);
4233 if (pri == LOW_PRIORITY)
4234 sched_prio(td, PRI_MAX_KERN);
4243 printf("ntoskrnl dummy called...\n");
4246 image_patch_table ntoskrnl_functbl[] = {
4247 IMPORT_SFUNC(RtlZeroMemory, 2),
4248 IMPORT_SFUNC(RtlSecureZeroMemory, 2),
4249 IMPORT_SFUNC(RtlFillMemory, 3),
4250 IMPORT_SFUNC(RtlMoveMemory, 3),
4251 IMPORT_SFUNC(RtlCharToInteger, 3),
4252 IMPORT_SFUNC(RtlCopyMemory, 3),
4253 IMPORT_SFUNC(RtlCopyString, 2),
4254 IMPORT_SFUNC(RtlCompareMemory, 3),
4255 IMPORT_SFUNC(RtlEqualUnicodeString, 3),
4256 IMPORT_SFUNC(RtlCopyUnicodeString, 2),
4257 IMPORT_SFUNC(RtlUnicodeStringToAnsiString, 3),
4258 IMPORT_SFUNC(RtlAnsiStringToUnicodeString, 3),
4259 IMPORT_SFUNC(RtlInitAnsiString, 2),
4260 IMPORT_SFUNC_MAP(RtlInitString, RtlInitAnsiString, 2),
4261 IMPORT_SFUNC(RtlInitUnicodeString, 2),
4262 IMPORT_SFUNC(RtlFreeAnsiString, 1),
4263 IMPORT_SFUNC(RtlFreeUnicodeString, 1),
4264 IMPORT_SFUNC(RtlUnicodeStringToInteger, 3),
4265 IMPORT_CFUNC(sprintf, 0),
4266 IMPORT_CFUNC(vsprintf, 0),
4267 IMPORT_CFUNC_MAP(_snprintf, snprintf, 0),
4268 IMPORT_CFUNC_MAP(_vsnprintf, vsnprintf, 0),
4269 IMPORT_CFUNC(DbgPrint, 0),
4270 IMPORT_SFUNC(DbgBreakPoint, 0),
4271 IMPORT_SFUNC(KeBugCheckEx, 5),
4272 IMPORT_CFUNC(strncmp, 0),
4273 IMPORT_CFUNC(strcmp, 0),
4274 IMPORT_CFUNC_MAP(stricmp, strcasecmp, 0),
4275 IMPORT_CFUNC(strncpy, 0),
4276 IMPORT_CFUNC(strcpy, 0),
4277 IMPORT_CFUNC(strlen, 0),
4278 IMPORT_CFUNC_MAP(toupper, ntoskrnl_toupper, 0),
4279 IMPORT_CFUNC_MAP(tolower, ntoskrnl_tolower, 0),
4280 IMPORT_CFUNC_MAP(strstr, ntoskrnl_strstr, 0),
4281 IMPORT_CFUNC_MAP(strncat, ntoskrnl_strncat, 0),
4282 IMPORT_CFUNC_MAP(strchr, index, 0),
4283 IMPORT_CFUNC_MAP(strrchr, rindex, 0),
4284 IMPORT_CFUNC(memcpy, 0),
4285 IMPORT_CFUNC_MAP(memmove, ntoskrnl_memmove, 0),
4286 IMPORT_CFUNC_MAP(memset, ntoskrnl_memset, 0),
4287 IMPORT_CFUNC_MAP(memchr, ntoskrnl_memchr, 0),
4288 IMPORT_SFUNC(IoAllocateDriverObjectExtension, 4),
4289 IMPORT_SFUNC(IoGetDriverObjectExtension, 2),
4290 IMPORT_FFUNC(IofCallDriver, 2),
4291 IMPORT_FFUNC(IofCompleteRequest, 2),
4292 IMPORT_SFUNC(IoAcquireCancelSpinLock, 1),
4293 IMPORT_SFUNC(IoReleaseCancelSpinLock, 1),
4294 IMPORT_SFUNC(IoCancelIrp, 1),
4295 IMPORT_SFUNC(IoConnectInterrupt, 11),
4296 IMPORT_SFUNC(IoDisconnectInterrupt, 1),
4297 IMPORT_SFUNC(IoCreateDevice, 7),
4298 IMPORT_SFUNC(IoDeleteDevice, 1),
4299 IMPORT_SFUNC(IoGetAttachedDevice, 1),
4300 IMPORT_SFUNC(IoAttachDeviceToDeviceStack, 2),
4301 IMPORT_SFUNC(IoDetachDevice, 1),
4302 IMPORT_SFUNC(IoBuildSynchronousFsdRequest, 7),
4303 IMPORT_SFUNC(IoBuildAsynchronousFsdRequest, 6),
4304 IMPORT_SFUNC(IoBuildDeviceIoControlRequest, 9),
4305 IMPORT_SFUNC(IoAllocateIrp, 2),
4306 IMPORT_SFUNC(IoReuseIrp, 2),
4307 IMPORT_SFUNC(IoMakeAssociatedIrp, 2),
4308 IMPORT_SFUNC(IoFreeIrp, 1),
4309 IMPORT_SFUNC(IoInitializeIrp, 3),
4310 IMPORT_SFUNC(KeAcquireInterruptSpinLock, 1),
4311 IMPORT_SFUNC(KeReleaseInterruptSpinLock, 2),
4312 IMPORT_SFUNC(KeSynchronizeExecution, 3),
4313 IMPORT_SFUNC(KeWaitForSingleObject, 5),
4314 IMPORT_SFUNC(KeWaitForMultipleObjects, 8),
4315 IMPORT_SFUNC(_allmul, 4),
4316 IMPORT_SFUNC(_alldiv, 4),
4317 IMPORT_SFUNC(_allrem, 4),
4318 IMPORT_RFUNC(_allshr, 0),
4319 IMPORT_RFUNC(_allshl, 0),
4320 IMPORT_SFUNC(_aullmul, 4),
4321 IMPORT_SFUNC(_aulldiv, 4),
4322 IMPORT_SFUNC(_aullrem, 4),
4323 IMPORT_RFUNC(_aullshr, 0),
4324 IMPORT_RFUNC(_aullshl, 0),
4325 IMPORT_CFUNC(atoi, 0),
4326 IMPORT_CFUNC(atol, 0),
4327 IMPORT_CFUNC(rand, 0),
4328 IMPORT_CFUNC(srand, 0),
4329 IMPORT_SFUNC(WRITE_REGISTER_USHORT, 2),
4330 IMPORT_SFUNC(READ_REGISTER_USHORT, 1),
4331 IMPORT_SFUNC(WRITE_REGISTER_ULONG, 2),
4332 IMPORT_SFUNC(READ_REGISTER_ULONG, 1),
4333 IMPORT_SFUNC(READ_REGISTER_UCHAR, 1),
4334 IMPORT_SFUNC(WRITE_REGISTER_UCHAR, 2),
4335 IMPORT_SFUNC(ExInitializePagedLookasideList, 7),
4336 IMPORT_SFUNC(ExDeletePagedLookasideList, 1),
4337 IMPORT_SFUNC(ExInitializeNPagedLookasideList, 7),
4338 IMPORT_SFUNC(ExDeleteNPagedLookasideList, 1),
4339 IMPORT_FFUNC(InterlockedPopEntrySList, 1),
4340 IMPORT_FFUNC(InitializeSListHead, 1),
4341 IMPORT_FFUNC(InterlockedPushEntrySList, 2),
4342 IMPORT_SFUNC(ExQueryDepthSList, 1),
4343 IMPORT_FFUNC_MAP(ExpInterlockedPopEntrySList,
4344 InterlockedPopEntrySList, 1),
4345 IMPORT_FFUNC_MAP(ExpInterlockedPushEntrySList,
4346 InterlockedPushEntrySList, 2),
4347 IMPORT_FFUNC(ExInterlockedPopEntrySList, 2),
4348 IMPORT_FFUNC(ExInterlockedPushEntrySList, 3),
4349 IMPORT_SFUNC(ExAllocatePoolWithTag, 3),
4350 IMPORT_SFUNC(ExFreePoolWithTag, 2),
4351 IMPORT_SFUNC(ExFreePool, 1),
4353 IMPORT_FFUNC(KefAcquireSpinLockAtDpcLevel, 1),
4354 IMPORT_FFUNC(KefReleaseSpinLockFromDpcLevel,1),
4355 IMPORT_FFUNC(KeAcquireSpinLockRaiseToDpc, 1),
4358 * For AMD64, we can get away with just mapping
4359 * KeAcquireSpinLockRaiseToDpc() directly to KfAcquireSpinLock()
4360 * because the calling conventions end up being the same.
4361 * On i386, we have to be careful because KfAcquireSpinLock()
4362 * is _fastcall but KeAcquireSpinLockRaiseToDpc() isn't.
4364 IMPORT_SFUNC(KeAcquireSpinLockAtDpcLevel, 1),
4365 IMPORT_SFUNC(KeReleaseSpinLockFromDpcLevel, 1),
4366 IMPORT_SFUNC_MAP(KeAcquireSpinLockRaiseToDpc, KfAcquireSpinLock, 1),
4368 IMPORT_SFUNC_MAP(KeReleaseSpinLock, KfReleaseSpinLock, 1),
4369 IMPORT_FFUNC(InterlockedIncrement, 1),
4370 IMPORT_FFUNC(InterlockedDecrement, 1),
4371 IMPORT_FFUNC(InterlockedExchange, 2),
4372 IMPORT_FFUNC(ExInterlockedAddLargeStatistic, 2),
4373 IMPORT_SFUNC(IoAllocateMdl, 5),
4374 IMPORT_SFUNC(IoFreeMdl, 1),
4375 IMPORT_SFUNC(MmAllocateContiguousMemory, 2 + 1),
4376 IMPORT_SFUNC(MmAllocateContiguousMemorySpecifyCache, 5 + 3),
4377 IMPORT_SFUNC(MmFreeContiguousMemory, 1),
4378 IMPORT_SFUNC(MmFreeContiguousMemorySpecifyCache, 3),
4379 IMPORT_SFUNC(MmSizeOfMdl, 1),
4380 IMPORT_SFUNC(MmMapLockedPages, 2),
4381 IMPORT_SFUNC(MmMapLockedPagesSpecifyCache, 6),
4382 IMPORT_SFUNC(MmUnmapLockedPages, 2),
4383 IMPORT_SFUNC(MmBuildMdlForNonPagedPool, 1),
4384 IMPORT_SFUNC(MmGetPhysicalAddress, 1),
4385 IMPORT_SFUNC(MmGetSystemRoutineAddress, 1),
4386 IMPORT_SFUNC(MmIsAddressValid, 1),
4387 IMPORT_SFUNC(MmMapIoSpace, 3 + 1),
4388 IMPORT_SFUNC(MmUnmapIoSpace, 2),
4389 IMPORT_SFUNC(KeInitializeSpinLock, 1),
4390 IMPORT_SFUNC(IoIsWdmVersionAvailable, 2),
4391 IMPORT_SFUNC(IoOpenDeviceRegistryKey, 4),
4392 IMPORT_SFUNC(IoGetDeviceObjectPointer, 4),
4393 IMPORT_SFUNC(IoGetDeviceProperty, 5),
4394 IMPORT_SFUNC(IoAllocateWorkItem, 1),
4395 IMPORT_SFUNC(IoFreeWorkItem, 1),
4396 IMPORT_SFUNC(IoQueueWorkItem, 4),
4397 IMPORT_SFUNC(ExQueueWorkItem, 2),
4398 IMPORT_SFUNC(ntoskrnl_workitem, 2),
4399 IMPORT_SFUNC(KeInitializeMutex, 2),
4400 IMPORT_SFUNC(KeReleaseMutex, 2),
4401 IMPORT_SFUNC(KeReadStateMutex, 1),
4402 IMPORT_SFUNC(KeInitializeEvent, 3),
4403 IMPORT_SFUNC(KeSetEvent, 3),
4404 IMPORT_SFUNC(KeResetEvent, 1),
4405 IMPORT_SFUNC(KeClearEvent, 1),
4406 IMPORT_SFUNC(KeReadStateEvent, 1),
4407 IMPORT_SFUNC(KeInitializeTimer, 1),
4408 IMPORT_SFUNC(KeInitializeTimerEx, 2),
4409 IMPORT_SFUNC(KeSetTimer, 3),
4410 IMPORT_SFUNC(KeSetTimerEx, 4),
4411 IMPORT_SFUNC(KeCancelTimer, 1),
4412 IMPORT_SFUNC(KeReadStateTimer, 1),
4413 IMPORT_SFUNC(KeInitializeDpc, 3),
4414 IMPORT_SFUNC(KeInsertQueueDpc, 3),
4415 IMPORT_SFUNC(KeRemoveQueueDpc, 1),
4416 IMPORT_SFUNC(KeSetImportanceDpc, 2),
4417 IMPORT_SFUNC(KeSetTargetProcessorDpc, 2),
4418 IMPORT_SFUNC(KeFlushQueuedDpcs, 0),
4419 IMPORT_SFUNC(KeGetCurrentProcessorNumber, 1),
4420 IMPORT_SFUNC(ObReferenceObjectByHandle, 6),
4421 IMPORT_FFUNC(ObfDereferenceObject, 1),
4422 IMPORT_SFUNC(ZwClose, 1),
4423 IMPORT_SFUNC(PsCreateSystemThread, 7),
4424 IMPORT_SFUNC(PsTerminateSystemThread, 1),
4425 IMPORT_SFUNC(IoWMIRegistrationControl, 2),
4426 IMPORT_SFUNC(WmiQueryTraceInformation, 5),
4427 IMPORT_CFUNC(WmiTraceMessage, 0),
4428 IMPORT_SFUNC(KeQuerySystemTime, 1),
4429 IMPORT_CFUNC(KeTickCount, 0),
4430 IMPORT_SFUNC(KeDelayExecutionThread, 3),
4431 IMPORT_SFUNC(KeQueryInterruptTime, 0),
4432 IMPORT_SFUNC(KeGetCurrentThread, 0),
4433 IMPORT_SFUNC(KeSetPriorityThread, 2),
4436 * This last entry is a catch-all for any function we haven't
4437 * implemented yet. The PE import list patching routine will
4438 * use it for any function that doesn't have an explicit match
4442 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
4445 { NULL, NULL, NULL }