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, CTLTYPE_INT | CTLFLAG_RW,
86 NULL, 0, sysctl_show_timers, "I",
87 "Show ntoskrnl timer stats");
101 typedef struct kdpc_queue kdpc_queue;
105 struct thread *we_td;
108 typedef struct wb_ext wb_ext;
110 #define NTOSKRNL_TIMEOUTS 256
111 #ifdef NTOSKRNL_DEBUG_TIMERS
112 static uint64_t ntoskrnl_timer_fires;
113 static uint64_t ntoskrnl_timer_sets;
114 static uint64_t ntoskrnl_timer_reloads;
115 static uint64_t ntoskrnl_timer_cancels;
118 struct callout_entry {
119 struct callout ce_callout;
123 typedef struct callout_entry callout_entry;
125 static struct list_entry ntoskrnl_calllist;
126 static struct mtx ntoskrnl_calllock;
127 struct kuser_shared_data kuser_shared_data;
129 static struct list_entry ntoskrnl_intlist;
130 static kspin_lock ntoskrnl_intlock;
132 static uint8_t RtlEqualUnicodeString(unicode_string *,
133 unicode_string *, uint8_t);
134 static void RtlCopyString(ansi_string *, const ansi_string *);
135 static void RtlCopyUnicodeString(unicode_string *,
137 static irp *IoBuildSynchronousFsdRequest(uint32_t, device_object *,
138 void *, uint32_t, uint64_t *, nt_kevent *, io_status_block *);
139 static irp *IoBuildAsynchronousFsdRequest(uint32_t,
140 device_object *, void *, uint32_t, uint64_t *, io_status_block *);
141 static irp *IoBuildDeviceIoControlRequest(uint32_t,
142 device_object *, void *, uint32_t, void *, uint32_t,
143 uint8_t, nt_kevent *, io_status_block *);
144 static irp *IoAllocateIrp(uint8_t, uint8_t);
145 static void IoReuseIrp(irp *, uint32_t);
146 static void IoFreeIrp(irp *);
147 static void IoInitializeIrp(irp *, uint16_t, uint8_t);
148 static irp *IoMakeAssociatedIrp(irp *, uint8_t);
149 static uint32_t KeWaitForMultipleObjects(uint32_t,
150 nt_dispatch_header **, uint32_t, uint32_t, uint32_t, uint8_t,
151 int64_t *, wait_block *);
152 static void ntoskrnl_waittest(nt_dispatch_header *, uint32_t);
153 static void ntoskrnl_satisfy_wait(nt_dispatch_header *, struct thread *);
154 static void ntoskrnl_satisfy_multiple_waits(wait_block *);
155 static int ntoskrnl_is_signalled(nt_dispatch_header *, struct thread *);
156 static void ntoskrnl_insert_timer(ktimer *, int);
157 static void ntoskrnl_remove_timer(ktimer *);
158 #ifdef NTOSKRNL_DEBUG_TIMERS
159 static void ntoskrnl_show_timers(void);
161 static void ntoskrnl_timercall(void *);
162 static void ntoskrnl_dpc_thread(void *);
163 static void ntoskrnl_destroy_dpc_threads(void);
164 static void ntoskrnl_destroy_workitem_threads(void);
165 static void ntoskrnl_workitem_thread(void *);
166 static void ntoskrnl_workitem(device_object *, void *);
167 static void ntoskrnl_unicode_to_ascii(uint16_t *, char *, int);
168 static void ntoskrnl_ascii_to_unicode(char *, uint16_t *, int);
169 static uint8_t ntoskrnl_insert_dpc(list_entry *, kdpc *);
170 static void WRITE_REGISTER_USHORT(uint16_t *, uint16_t);
171 static uint16_t READ_REGISTER_USHORT(uint16_t *);
172 static void WRITE_REGISTER_ULONG(uint32_t *, uint32_t);
173 static uint32_t READ_REGISTER_ULONG(uint32_t *);
174 static void WRITE_REGISTER_UCHAR(uint8_t *, uint8_t);
175 static uint8_t READ_REGISTER_UCHAR(uint8_t *);
176 static int64_t _allmul(int64_t, int64_t);
177 static int64_t _alldiv(int64_t, int64_t);
178 static int64_t _allrem(int64_t, int64_t);
179 static int64_t _allshr(int64_t, uint8_t);
180 static int64_t _allshl(int64_t, uint8_t);
181 static uint64_t _aullmul(uint64_t, uint64_t);
182 static uint64_t _aulldiv(uint64_t, uint64_t);
183 static uint64_t _aullrem(uint64_t, uint64_t);
184 static uint64_t _aullshr(uint64_t, uint8_t);
185 static uint64_t _aullshl(uint64_t, uint8_t);
186 static slist_entry *ntoskrnl_pushsl(slist_header *, slist_entry *);
187 static void InitializeSListHead(slist_header *);
188 static slist_entry *ntoskrnl_popsl(slist_header *);
189 static void ExFreePoolWithTag(void *, uint32_t);
190 static void ExInitializePagedLookasideList(paged_lookaside_list *,
191 lookaside_alloc_func *, lookaside_free_func *,
192 uint32_t, size_t, uint32_t, uint16_t);
193 static void ExDeletePagedLookasideList(paged_lookaside_list *);
194 static void ExInitializeNPagedLookasideList(npaged_lookaside_list *,
195 lookaside_alloc_func *, lookaside_free_func *,
196 uint32_t, size_t, uint32_t, uint16_t);
197 static void ExDeleteNPagedLookasideList(npaged_lookaside_list *);
199 *ExInterlockedPushEntrySList(slist_header *,
200 slist_entry *, kspin_lock *);
202 *ExInterlockedPopEntrySList(slist_header *, kspin_lock *);
203 static uint32_t InterlockedIncrement(volatile uint32_t *);
204 static uint32_t InterlockedDecrement(volatile uint32_t *);
205 static void ExInterlockedAddLargeStatistic(uint64_t *, uint32_t);
206 static void *MmAllocateContiguousMemory(uint32_t, uint64_t);
207 static void *MmAllocateContiguousMemorySpecifyCache(uint32_t,
208 uint64_t, uint64_t, uint64_t, enum nt_caching_type);
209 static void MmFreeContiguousMemory(void *);
210 static void MmFreeContiguousMemorySpecifyCache(void *, uint32_t,
211 enum nt_caching_type);
212 static uint32_t MmSizeOfMdl(void *, size_t);
213 static void *MmMapLockedPages(mdl *, uint8_t);
214 static void *MmMapLockedPagesSpecifyCache(mdl *,
215 uint8_t, uint32_t, void *, uint32_t, uint32_t);
216 static void MmUnmapLockedPages(void *, mdl *);
217 static device_t ntoskrnl_finddev(device_t, uint64_t, struct resource **);
218 static void RtlZeroMemory(void *, size_t);
219 static void RtlSecureZeroMemory(void *, size_t);
220 static void RtlFillMemory(void *, size_t, uint8_t);
221 static void RtlMoveMemory(void *, const void *, size_t);
222 static ndis_status RtlCharToInteger(const char *, uint32_t, uint32_t *);
223 static void RtlCopyMemory(void *, const void *, size_t);
224 static size_t RtlCompareMemory(const void *, const void *, size_t);
225 static ndis_status RtlUnicodeStringToInteger(unicode_string *,
226 uint32_t, uint32_t *);
227 static int atoi (const char *);
228 static long atol (const char *);
229 static int rand(void);
230 static void srand(unsigned int);
231 static void KeQuerySystemTime(uint64_t *);
232 static uint32_t KeTickCount(void);
233 static uint8_t IoIsWdmVersionAvailable(uint8_t, uint8_t);
234 static int32_t IoOpenDeviceRegistryKey(struct device_object *, uint32_t,
236 static void ntoskrnl_thrfunc(void *);
237 static ndis_status PsCreateSystemThread(ndis_handle *,
238 uint32_t, void *, ndis_handle, void *, void *, void *);
239 static ndis_status PsTerminateSystemThread(ndis_status);
240 static ndis_status IoGetDeviceObjectPointer(unicode_string *,
241 uint32_t, void *, device_object *);
242 static ndis_status IoGetDeviceProperty(device_object *, uint32_t,
243 uint32_t, void *, uint32_t *);
244 static void KeInitializeMutex(kmutant *, uint32_t);
245 static uint32_t KeReleaseMutex(kmutant *, uint8_t);
246 static uint32_t KeReadStateMutex(kmutant *);
247 static ndis_status ObReferenceObjectByHandle(ndis_handle,
248 uint32_t, void *, uint8_t, void **, void **);
249 static void ObfDereferenceObject(void *);
250 static uint32_t ZwClose(ndis_handle);
251 static uint32_t WmiQueryTraceInformation(uint32_t, void *, uint32_t,
253 static uint32_t WmiTraceMessage(uint64_t, uint32_t, void *, uint16_t, ...);
254 static uint32_t IoWMIRegistrationControl(device_object *, uint32_t);
255 static void *ntoskrnl_memset(void *, int, size_t);
256 static void *ntoskrnl_memmove(void *, void *, size_t);
257 static void *ntoskrnl_memchr(void *, unsigned char, size_t);
258 static char *ntoskrnl_strstr(char *, char *);
259 static char *ntoskrnl_strncat(char *, char *, size_t);
260 static int ntoskrnl_toupper(int);
261 static int ntoskrnl_tolower(int);
262 static funcptr ntoskrnl_findwrap(funcptr);
263 static uint32_t DbgPrint(char *, ...);
264 static void DbgBreakPoint(void);
265 static void KeBugCheckEx(uint32_t, u_long, u_long, u_long, u_long);
266 static int32_t KeDelayExecutionThread(uint8_t, uint8_t, int64_t *);
267 static int32_t KeSetPriorityThread(struct thread *, int32_t);
268 static void dummy(void);
270 static struct mtx ntoskrnl_dispatchlock;
271 static struct mtx ntoskrnl_interlock;
272 static kspin_lock ntoskrnl_cancellock;
273 static int ntoskrnl_kth = 0;
274 static struct nt_objref_head ntoskrnl_reflist;
275 static uma_zone_t mdl_zone;
276 static uma_zone_t iw_zone;
277 static struct kdpc_queue *kq_queues;
278 static struct kdpc_queue *wq_queues;
279 static int wq_idx = 0;
284 image_patch_table *patch;
291 mtx_init(&ntoskrnl_dispatchlock,
292 "ntoskrnl dispatch lock", MTX_NDIS_LOCK, MTX_DEF|MTX_RECURSE);
293 mtx_init(&ntoskrnl_interlock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
294 KeInitializeSpinLock(&ntoskrnl_cancellock);
295 KeInitializeSpinLock(&ntoskrnl_intlock);
296 TAILQ_INIT(&ntoskrnl_reflist);
298 InitializeListHead(&ntoskrnl_calllist);
299 InitializeListHead(&ntoskrnl_intlist);
300 mtx_init(&ntoskrnl_calllock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
302 kq_queues = ExAllocatePoolWithTag(NonPagedPool,
303 #ifdef NTOSKRNL_MULTIPLE_DPCS
304 sizeof(kdpc_queue) * mp_ncpus, 0);
306 sizeof(kdpc_queue), 0);
309 if (kq_queues == NULL)
312 wq_queues = ExAllocatePoolWithTag(NonPagedPool,
313 sizeof(kdpc_queue) * WORKITEM_THREADS, 0);
315 if (wq_queues == NULL)
318 #ifdef NTOSKRNL_MULTIPLE_DPCS
319 bzero((char *)kq_queues, sizeof(kdpc_queue) * mp_ncpus);
321 bzero((char *)kq_queues, sizeof(kdpc_queue));
323 bzero((char *)wq_queues, sizeof(kdpc_queue) * WORKITEM_THREADS);
326 * Launch the DPC threads.
329 #ifdef NTOSKRNL_MULTIPLE_DPCS
330 for (i = 0; i < mp_ncpus; i++) {
332 for (i = 0; i < 1; i++) {
336 error = kproc_create(ntoskrnl_dpc_thread, kq, &p,
337 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows DPC %d", i);
339 panic("failed to launch DPC thread");
343 * Launch the workitem threads.
346 for (i = 0; i < WORKITEM_THREADS; i++) {
348 error = kproc_create(ntoskrnl_workitem_thread, kq, &p,
349 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Workitem %d", i);
351 panic("failed to launch workitem thread");
354 patch = ntoskrnl_functbl;
355 while (patch->ipt_func != NULL) {
356 windrv_wrap((funcptr)patch->ipt_func,
357 (funcptr *)&patch->ipt_wrap,
358 patch->ipt_argcnt, patch->ipt_ftype);
362 for (i = 0; i < NTOSKRNL_TIMEOUTS; i++) {
363 e = ExAllocatePoolWithTag(NonPagedPool,
364 sizeof(callout_entry), 0);
366 panic("failed to allocate timeouts");
367 mtx_lock_spin(&ntoskrnl_calllock);
368 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
369 mtx_unlock_spin(&ntoskrnl_calllock);
373 * MDLs are supposed to be variable size (they describe
374 * buffers containing some number of pages, but we don't
375 * know ahead of time how many pages that will be). But
376 * always allocating them off the heap is very slow. As
377 * a compromise, we create an MDL UMA zone big enough to
378 * handle any buffer requiring up to 16 pages, and we
379 * use those for any MDLs for buffers of 16 pages or less
380 * in size. For buffers larger than that (which we assume
381 * will be few and far between, we allocate the MDLs off
385 mdl_zone = uma_zcreate("Windows MDL", MDL_ZONE_SIZE,
386 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
388 iw_zone = uma_zcreate("Windows WorkItem", sizeof(io_workitem),
389 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
397 image_patch_table *patch;
401 patch = ntoskrnl_functbl;
402 while (patch->ipt_func != NULL) {
403 windrv_unwrap(patch->ipt_wrap);
407 /* Stop the workitem queues. */
408 ntoskrnl_destroy_workitem_threads();
409 /* Stop the DPC queues. */
410 ntoskrnl_destroy_dpc_threads();
412 ExFreePool(kq_queues);
413 ExFreePool(wq_queues);
415 uma_zdestroy(mdl_zone);
416 uma_zdestroy(iw_zone);
418 mtx_lock_spin(&ntoskrnl_calllock);
419 while(!IsListEmpty(&ntoskrnl_calllist)) {
420 l = RemoveHeadList(&ntoskrnl_calllist);
421 e = CONTAINING_RECORD(l, callout_entry, ce_list);
422 mtx_unlock_spin(&ntoskrnl_calllock);
424 mtx_lock_spin(&ntoskrnl_calllock);
426 mtx_unlock_spin(&ntoskrnl_calllock);
428 mtx_destroy(&ntoskrnl_dispatchlock);
429 mtx_destroy(&ntoskrnl_interlock);
430 mtx_destroy(&ntoskrnl_calllock);
436 * We need to be able to reference this externally from the wrapper;
437 * GCC only generates a local implementation of memset.
440 ntoskrnl_memset(buf, ch, size)
445 return (memset(buf, ch, size));
449 ntoskrnl_memmove(dst, src, size)
454 bcopy(src, dst, size);
459 ntoskrnl_memchr(void *buf, unsigned char ch, size_t len)
462 unsigned char *p = buf;
467 } while (--len != 0);
473 ntoskrnl_strstr(s, find)
479 if ((c = *find++) != 0) {
483 if ((sc = *s++) == 0)
486 } while (strncmp(s, find, len) != 0);
492 /* Taken from libc */
494 ntoskrnl_strncat(dst, src, n)
506 if ((*d = *s++) == 0)
530 RtlEqualUnicodeString(unicode_string *str1, unicode_string *str2,
531 uint8_t caseinsensitive)
535 if (str1->us_len != str2->us_len)
538 for (i = 0; i < str1->us_len; i++) {
539 if (caseinsensitive == TRUE) {
540 if (toupper((char)(str1->us_buf[i] & 0xFF)) !=
541 toupper((char)(str2->us_buf[i] & 0xFF)))
544 if (str1->us_buf[i] != str2->us_buf[i])
553 RtlCopyString(dst, src)
555 const ansi_string *src;
557 if (src != NULL && src->as_buf != NULL && dst->as_buf != NULL) {
558 dst->as_len = min(src->as_len, dst->as_maxlen);
559 memcpy(dst->as_buf, src->as_buf, dst->as_len);
560 if (dst->as_len < dst->as_maxlen)
561 dst->as_buf[dst->as_len] = 0;
567 RtlCopyUnicodeString(dest, src)
568 unicode_string *dest;
572 if (dest->us_maxlen >= src->us_len)
573 dest->us_len = src->us_len;
575 dest->us_len = dest->us_maxlen;
576 memcpy(dest->us_buf, src->us_buf, dest->us_len);
580 ntoskrnl_ascii_to_unicode(ascii, unicode, len)
589 for (i = 0; i < len; i++) {
590 *ustr = (uint16_t)ascii[i];
596 ntoskrnl_unicode_to_ascii(unicode, ascii, len)
605 for (i = 0; i < len / 2; i++) {
606 *astr = (uint8_t)unicode[i];
612 RtlUnicodeStringToAnsiString(ansi_string *dest, unicode_string *src, uint8_t allocate)
614 if (dest == NULL || src == NULL)
615 return (STATUS_INVALID_PARAMETER);
617 dest->as_len = src->us_len / 2;
618 if (dest->as_maxlen < dest->as_len)
619 dest->as_len = dest->as_maxlen;
621 if (allocate == TRUE) {
622 dest->as_buf = ExAllocatePoolWithTag(NonPagedPool,
623 (src->us_len / 2) + 1, 0);
624 if (dest->as_buf == NULL)
625 return (STATUS_INSUFFICIENT_RESOURCES);
626 dest->as_len = dest->as_maxlen = src->us_len / 2;
628 dest->as_len = src->us_len / 2; /* XXX */
629 if (dest->as_maxlen < dest->as_len)
630 dest->as_len = dest->as_maxlen;
633 ntoskrnl_unicode_to_ascii(src->us_buf, dest->as_buf,
636 return (STATUS_SUCCESS);
640 RtlAnsiStringToUnicodeString(unicode_string *dest, ansi_string *src,
643 if (dest == NULL || src == NULL)
644 return (STATUS_INVALID_PARAMETER);
646 if (allocate == TRUE) {
647 dest->us_buf = ExAllocatePoolWithTag(NonPagedPool,
649 if (dest->us_buf == NULL)
650 return (STATUS_INSUFFICIENT_RESOURCES);
651 dest->us_len = dest->us_maxlen = strlen(src->as_buf) * 2;
653 dest->us_len = src->as_len * 2; /* XXX */
654 if (dest->us_maxlen < dest->us_len)
655 dest->us_len = dest->us_maxlen;
658 ntoskrnl_ascii_to_unicode(src->as_buf, dest->us_buf,
661 return (STATUS_SUCCESS);
665 ExAllocatePoolWithTag(pooltype, len, tag)
672 buf = malloc(len, M_DEVBUF, M_NOWAIT|M_ZERO);
680 ExFreePoolWithTag(buf, tag)
695 IoAllocateDriverObjectExtension(drv, clid, extlen, ext)
701 custom_extension *ce;
703 ce = ExAllocatePoolWithTag(NonPagedPool, sizeof(custom_extension)
707 return (STATUS_INSUFFICIENT_RESOURCES);
710 InsertTailList((&drv->dro_driverext->dre_usrext), (&ce->ce_list));
712 *ext = (void *)(ce + 1);
714 return (STATUS_SUCCESS);
718 IoGetDriverObjectExtension(drv, clid)
723 custom_extension *ce;
726 * Sanity check. Our dummy bus drivers don't have
727 * any driver extensions.
730 if (drv->dro_driverext == NULL)
733 e = drv->dro_driverext->dre_usrext.nle_flink;
734 while (e != &drv->dro_driverext->dre_usrext) {
735 ce = (custom_extension *)e;
736 if (ce->ce_clid == clid)
737 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);
1603 * KeWaitForSingleObject() is a tricky beast, because it can be used
1604 * with several different object types: semaphores, timers, events,
1605 * mutexes and threads. Semaphores don't appear very often, but the
1606 * other object types are quite common. KeWaitForSingleObject() is
1607 * what's normally used to acquire a mutex, and it can be used to
1608 * wait for a thread termination.
1610 * The Windows NDIS API is implemented in terms of Windows kernel
1611 * primitives, and some of the object manipulation is duplicated in
1612 * NDIS. For example, NDIS has timers and events, which are actually
1613 * Windows kevents and ktimers. Now, you're supposed to only use the
1614 * NDIS variants of these objects within the confines of the NDIS API,
1615 * but there are some naughty developers out there who will use
1616 * KeWaitForSingleObject() on NDIS timer and event objects, so we
1617 * have to support that as well. Conseqently, our NDIS timer and event
1618 * code has to be closely tied into our ntoskrnl timer and event code,
1619 * just as it is in Windows.
1621 * KeWaitForSingleObject() may do different things for different kinds
1624 * - For events, we check if the event has been signalled. If the
1625 * event is already in the signalled state, we just return immediately,
1626 * otherwise we wait for it to be set to the signalled state by someone
1627 * else calling KeSetEvent(). Events can be either synchronization or
1628 * notification events.
1630 * - For timers, if the timer has already fired and the timer is in
1631 * the signalled state, we just return, otherwise we wait on the
1632 * timer. Unlike an event, timers get signalled automatically when
1633 * they expire rather than someone having to trip them manually.
1634 * Timers initialized with KeInitializeTimer() are always notification
1635 * events: KeInitializeTimerEx() lets you initialize a timer as
1636 * either a notification or synchronization event.
1638 * - For mutexes, we try to acquire the mutex and if we can't, we wait
1639 * on the mutex until it's available and then grab it. When a mutex is
1640 * released, it enters the signalled state, which wakes up one of the
1641 * threads waiting to acquire it. Mutexes are always synchronization
1644 * - For threads, the only thing we do is wait until the thread object
1645 * enters a signalled state, which occurs when the thread terminates.
1646 * Threads are always notification events.
1648 * A notification event wakes up all threads waiting on an object. A
1649 * synchronization event wakes up just one. Also, a synchronization event
1650 * is auto-clearing, which means we automatically set the event back to
1651 * the non-signalled state once the wakeup is done.
1655 KeWaitForSingleObject(void *arg, uint32_t reason, uint32_t mode,
1656 uint8_t alertable, int64_t *duetime)
1659 struct thread *td = curthread;
1664 nt_dispatch_header *obj;
1669 return (STATUS_INVALID_PARAMETER);
1671 mtx_lock(&ntoskrnl_dispatchlock);
1673 cv_init(&we.we_cv, "KeWFS");
1677 * Check to see if this object is already signalled,
1678 * and just return without waiting if it is.
1680 if (ntoskrnl_is_signalled(obj, td) == TRUE) {
1681 /* Sanity check the signal state value. */
1682 if (obj->dh_sigstate != INT32_MIN) {
1683 ntoskrnl_satisfy_wait(obj, curthread);
1684 mtx_unlock(&ntoskrnl_dispatchlock);
1685 return (STATUS_SUCCESS);
1688 * There's a limit to how many times we can
1689 * recursively acquire a mutant. If we hit
1690 * the limit, something is very wrong.
1692 if (obj->dh_type == DISP_TYPE_MUTANT) {
1693 mtx_unlock(&ntoskrnl_dispatchlock);
1694 panic("mutant limit exceeded");
1699 bzero((char *)&w, sizeof(wait_block));
1702 w.wb_waittype = WAITTYPE_ANY;
1705 w.wb_awakened = FALSE;
1706 w.wb_oldpri = td->td_priority;
1708 InsertTailList((&obj->dh_waitlisthead), (&w.wb_waitlist));
1711 * The timeout value is specified in 100 nanosecond units
1712 * and can be a positive or negative number. If it's positive,
1713 * then the duetime is absolute, and we need to convert it
1714 * to an absolute offset relative to now in order to use it.
1715 * If it's negative, then the duetime is relative and we
1716 * just have to convert the units.
1719 if (duetime != NULL) {
1721 tv.tv_sec = - (*duetime) / 10000000;
1722 tv.tv_usec = (- (*duetime) / 10) -
1723 (tv.tv_sec * 1000000);
1725 ntoskrnl_time(&curtime);
1726 if (*duetime < curtime)
1727 tv.tv_sec = tv.tv_usec = 0;
1729 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1730 tv.tv_usec = ((*duetime) - curtime) / 10 -
1731 (tv.tv_sec * 1000000);
1736 if (duetime == NULL)
1737 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1739 error = cv_timedwait(&we.we_cv,
1740 &ntoskrnl_dispatchlock, tvtohz(&tv));
1742 RemoveEntryList(&w.wb_waitlist);
1744 cv_destroy(&we.we_cv);
1746 /* We timed out. Leave the object alone and return status. */
1748 if (error == EWOULDBLOCK) {
1749 mtx_unlock(&ntoskrnl_dispatchlock);
1750 return (STATUS_TIMEOUT);
1753 mtx_unlock(&ntoskrnl_dispatchlock);
1755 return (STATUS_SUCCESS);
1757 return (KeWaitForMultipleObjects(1, &obj, WAITTYPE_ALL, reason,
1758 mode, alertable, duetime, &w));
1763 KeWaitForMultipleObjects(uint32_t cnt, nt_dispatch_header *obj[], uint32_t wtype,
1764 uint32_t reason, uint32_t mode, uint8_t alertable, int64_t *duetime,
1765 wait_block *wb_array)
1767 struct thread *td = curthread;
1768 wait_block *whead, *w;
1769 wait_block _wb_array[MAX_WAIT_OBJECTS];
1770 nt_dispatch_header *cur;
1772 int i, wcnt = 0, error = 0;
1774 struct timespec t1, t2;
1775 uint32_t status = STATUS_SUCCESS;
1778 if (cnt > MAX_WAIT_OBJECTS)
1779 return (STATUS_INVALID_PARAMETER);
1780 if (cnt > THREAD_WAIT_OBJECTS && wb_array == NULL)
1781 return (STATUS_INVALID_PARAMETER);
1783 mtx_lock(&ntoskrnl_dispatchlock);
1785 cv_init(&we.we_cv, "KeWFM");
1788 if (wb_array == NULL)
1793 bzero((char *)whead, sizeof(wait_block) * cnt);
1795 /* First pass: see if we can satisfy any waits immediately. */
1800 for (i = 0; i < cnt; i++) {
1801 InsertTailList((&obj[i]->dh_waitlisthead),
1804 w->wb_object = obj[i];
1805 w->wb_waittype = wtype;
1807 w->wb_awakened = FALSE;
1808 w->wb_oldpri = td->td_priority;
1812 if (ntoskrnl_is_signalled(obj[i], td)) {
1814 * There's a limit to how many times
1815 * we can recursively acquire a mutant.
1816 * If we hit the limit, something
1819 if (obj[i]->dh_sigstate == INT32_MIN &&
1820 obj[i]->dh_type == DISP_TYPE_MUTANT) {
1821 mtx_unlock(&ntoskrnl_dispatchlock);
1822 panic("mutant limit exceeded");
1826 * If this is a WAITTYPE_ANY wait, then
1827 * satisfy the waited object and exit
1831 if (wtype == WAITTYPE_ANY) {
1832 ntoskrnl_satisfy_wait(obj[i], td);
1833 status = STATUS_WAIT_0 + i;
1838 w->wb_object = NULL;
1839 RemoveEntryList(&w->wb_waitlist);
1845 * If this is a WAITTYPE_ALL wait and all objects are
1846 * already signalled, satisfy the waits and exit now.
1849 if (wtype == WAITTYPE_ALL && wcnt == 0) {
1850 for (i = 0; i < cnt; i++)
1851 ntoskrnl_satisfy_wait(obj[i], td);
1852 status = STATUS_SUCCESS;
1857 * Create a circular waitblock list. The waitcount
1858 * must always be non-zero when we get here.
1861 (w - 1)->wb_next = whead;
1863 /* Wait on any objects that aren't yet signalled. */
1865 /* Calculate timeout, if any. */
1867 if (duetime != NULL) {
1869 tv.tv_sec = - (*duetime) / 10000000;
1870 tv.tv_usec = (- (*duetime) / 10) -
1871 (tv.tv_sec * 1000000);
1873 ntoskrnl_time(&curtime);
1874 if (*duetime < curtime)
1875 tv.tv_sec = tv.tv_usec = 0;
1877 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1878 tv.tv_usec = ((*duetime) - curtime) / 10 -
1879 (tv.tv_sec * 1000000);
1887 if (duetime == NULL)
1888 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1890 error = cv_timedwait(&we.we_cv,
1891 &ntoskrnl_dispatchlock, tvtohz(&tv));
1893 /* Wait with timeout expired. */
1896 status = STATUS_TIMEOUT;
1902 /* See what's been signalled. */
1907 if (ntoskrnl_is_signalled(cur, td) == TRUE ||
1908 w->wb_awakened == TRUE) {
1909 /* Sanity check the signal state value. */
1910 if (cur->dh_sigstate == INT32_MIN &&
1911 cur->dh_type == DISP_TYPE_MUTANT) {
1912 mtx_unlock(&ntoskrnl_dispatchlock);
1913 panic("mutant limit exceeded");
1916 if (wtype == WAITTYPE_ANY) {
1917 status = w->wb_waitkey &
1923 } while (w != whead);
1926 * If all objects have been signalled, or if this
1927 * is a WAITTYPE_ANY wait and we were woke up by
1928 * someone, we can bail.
1932 status = STATUS_SUCCESS;
1937 * If this is WAITTYPE_ALL wait, and there's still
1938 * objects that haven't been signalled, deduct the
1939 * time that's elapsed so far from the timeout and
1940 * wait again (or continue waiting indefinitely if
1941 * there's no timeout).
1944 if (duetime != NULL) {
1945 tv.tv_sec -= (t2.tv_sec - t1.tv_sec);
1946 tv.tv_usec -= (t2.tv_nsec - t1.tv_nsec) / 1000;
1953 cv_destroy(&we.we_cv);
1955 for (i = 0; i < cnt; i++) {
1956 if (whead[i].wb_object != NULL)
1957 RemoveEntryList(&whead[i].wb_waitlist);
1960 mtx_unlock(&ntoskrnl_dispatchlock);
1966 WRITE_REGISTER_USHORT(uint16_t *reg, uint16_t val)
1968 bus_space_write_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1972 READ_REGISTER_USHORT(reg)
1975 return (bus_space_read_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1979 WRITE_REGISTER_ULONG(reg, val)
1983 bus_space_write_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1987 READ_REGISTER_ULONG(reg)
1990 return (bus_space_read_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1994 READ_REGISTER_UCHAR(uint8_t *reg)
1996 return (bus_space_read_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
2000 WRITE_REGISTER_UCHAR(uint8_t *reg, uint8_t val)
2002 bus_space_write_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
2054 _allshl(int64_t a, uint8_t b)
2060 _aullshl(uint64_t a, uint8_t b)
2066 _allshr(int64_t a, uint8_t b)
2072 _aullshr(uint64_t a, uint8_t b)
2077 static slist_entry *
2078 ntoskrnl_pushsl(head, entry)
2082 slist_entry *oldhead;
2084 oldhead = head->slh_list.slh_next;
2085 entry->sl_next = head->slh_list.slh_next;
2086 head->slh_list.slh_next = entry;
2087 head->slh_list.slh_depth++;
2088 head->slh_list.slh_seq++;
2094 InitializeSListHead(head)
2097 memset(head, 0, sizeof(*head));
2100 static slist_entry *
2101 ntoskrnl_popsl(head)
2106 first = head->slh_list.slh_next;
2107 if (first != NULL) {
2108 head->slh_list.slh_next = first->sl_next;
2109 head->slh_list.slh_depth--;
2110 head->slh_list.slh_seq++;
2117 * We need this to make lookaside lists work for amd64.
2118 * We pass a pointer to ExAllocatePoolWithTag() the lookaside
2119 * list structure. For amd64 to work right, this has to be a
2120 * pointer to the wrapped version of the routine, not the
2121 * original. Letting the Windows driver invoke the original
2122 * function directly will result in a convention calling
2123 * mismatch and a pretty crash. On x86, this effectively
2124 * becomes a no-op since ipt_func and ipt_wrap are the same.
2128 ntoskrnl_findwrap(func)
2131 image_patch_table *patch;
2133 patch = ntoskrnl_functbl;
2134 while (patch->ipt_func != NULL) {
2135 if ((funcptr)patch->ipt_func == func)
2136 return ((funcptr)patch->ipt_wrap);
2144 ExInitializePagedLookasideList(paged_lookaside_list *lookaside,
2145 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2146 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2148 bzero((char *)lookaside, sizeof(paged_lookaside_list));
2150 if (size < sizeof(slist_entry))
2151 lookaside->nll_l.gl_size = sizeof(slist_entry);
2153 lookaside->nll_l.gl_size = size;
2154 lookaside->nll_l.gl_tag = tag;
2155 if (allocfunc == NULL)
2156 lookaside->nll_l.gl_allocfunc =
2157 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2159 lookaside->nll_l.gl_allocfunc = allocfunc;
2161 if (freefunc == NULL)
2162 lookaside->nll_l.gl_freefunc =
2163 ntoskrnl_findwrap((funcptr)ExFreePool);
2165 lookaside->nll_l.gl_freefunc = freefunc;
2168 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2171 lookaside->nll_l.gl_type = NonPagedPool;
2172 lookaside->nll_l.gl_depth = depth;
2173 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2177 ExDeletePagedLookasideList(lookaside)
2178 paged_lookaside_list *lookaside;
2181 void (*freefunc)(void *);
2183 freefunc = lookaside->nll_l.gl_freefunc;
2184 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2185 MSCALL1(freefunc, buf);
2189 ExInitializeNPagedLookasideList(npaged_lookaside_list *lookaside,
2190 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2191 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2193 bzero((char *)lookaside, sizeof(npaged_lookaside_list));
2195 if (size < sizeof(slist_entry))
2196 lookaside->nll_l.gl_size = sizeof(slist_entry);
2198 lookaside->nll_l.gl_size = size;
2199 lookaside->nll_l.gl_tag = tag;
2200 if (allocfunc == NULL)
2201 lookaside->nll_l.gl_allocfunc =
2202 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2204 lookaside->nll_l.gl_allocfunc = allocfunc;
2206 if (freefunc == NULL)
2207 lookaside->nll_l.gl_freefunc =
2208 ntoskrnl_findwrap((funcptr)ExFreePool);
2210 lookaside->nll_l.gl_freefunc = freefunc;
2213 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2216 lookaside->nll_l.gl_type = NonPagedPool;
2217 lookaside->nll_l.gl_depth = depth;
2218 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2222 ExDeleteNPagedLookasideList(lookaside)
2223 npaged_lookaside_list *lookaside;
2226 void (*freefunc)(void *);
2228 freefunc = lookaside->nll_l.gl_freefunc;
2229 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2230 MSCALL1(freefunc, buf);
2234 InterlockedPushEntrySList(head, entry)
2238 slist_entry *oldhead;
2240 mtx_lock_spin(&ntoskrnl_interlock);
2241 oldhead = ntoskrnl_pushsl(head, entry);
2242 mtx_unlock_spin(&ntoskrnl_interlock);
2248 InterlockedPopEntrySList(head)
2253 mtx_lock_spin(&ntoskrnl_interlock);
2254 first = ntoskrnl_popsl(head);
2255 mtx_unlock_spin(&ntoskrnl_interlock);
2260 static slist_entry *
2261 ExInterlockedPushEntrySList(head, entry, lock)
2266 return (InterlockedPushEntrySList(head, entry));
2269 static slist_entry *
2270 ExInterlockedPopEntrySList(head, lock)
2274 return (InterlockedPopEntrySList(head));
2278 ExQueryDepthSList(head)
2283 mtx_lock_spin(&ntoskrnl_interlock);
2284 depth = head->slh_list.slh_depth;
2285 mtx_unlock_spin(&ntoskrnl_interlock);
2291 KeInitializeSpinLock(lock)
2299 KefAcquireSpinLockAtDpcLevel(lock)
2302 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2306 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0) {
2308 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2317 KefReleaseSpinLockFromDpcLevel(lock)
2320 atomic_store_rel_int((volatile u_int *)lock, 0);
2324 KeAcquireSpinLockRaiseToDpc(kspin_lock *lock)
2328 if (KeGetCurrentIrql() > DISPATCH_LEVEL)
2329 panic("IRQL_NOT_LESS_THAN_OR_EQUAL");
2331 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
2332 KeAcquireSpinLockAtDpcLevel(lock);
2338 KeAcquireSpinLockAtDpcLevel(kspin_lock *lock)
2340 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0)
2345 KeReleaseSpinLockFromDpcLevel(kspin_lock *lock)
2347 atomic_store_rel_int((volatile u_int *)lock, 0);
2349 #endif /* __i386__ */
2352 InterlockedExchange(dst, val)
2353 volatile uint32_t *dst;
2358 mtx_lock_spin(&ntoskrnl_interlock);
2361 mtx_unlock_spin(&ntoskrnl_interlock);
2367 InterlockedIncrement(addend)
2368 volatile uint32_t *addend;
2370 atomic_add_long((volatile u_long *)addend, 1);
2375 InterlockedDecrement(addend)
2376 volatile uint32_t *addend;
2378 atomic_subtract_long((volatile u_long *)addend, 1);
2383 ExInterlockedAddLargeStatistic(addend, inc)
2387 mtx_lock_spin(&ntoskrnl_interlock);
2389 mtx_unlock_spin(&ntoskrnl_interlock);
2393 IoAllocateMdl(void *vaddr, uint32_t len, uint8_t secondarybuf,
2394 uint8_t chargequota, irp *iopkt)
2399 if (MmSizeOfMdl(vaddr, len) > MDL_ZONE_SIZE)
2400 m = ExAllocatePoolWithTag(NonPagedPool,
2401 MmSizeOfMdl(vaddr, len), 0);
2403 m = uma_zalloc(mdl_zone, M_NOWAIT | M_ZERO);
2410 MmInitializeMdl(m, vaddr, len);
2413 * MmInitializMdl() clears the flags field, so we
2414 * have to set this here. If the MDL came from the
2415 * MDL UMA zone, tag it so we can release it to
2416 * the right place later.
2419 m->mdl_flags = MDL_ZONE_ALLOCED;
2421 if (iopkt != NULL) {
2422 if (secondarybuf == TRUE) {
2424 last = iopkt->irp_mdl;
2425 while (last->mdl_next != NULL)
2426 last = last->mdl_next;
2429 if (iopkt->irp_mdl != NULL)
2430 panic("leaking an MDL in IoAllocateMdl()");
2445 if (m->mdl_flags & MDL_ZONE_ALLOCED)
2446 uma_zfree(mdl_zone, m);
2452 MmAllocateContiguousMemory(size, highest)
2457 size_t pagelength = roundup(size, PAGE_SIZE);
2459 addr = ExAllocatePoolWithTag(NonPagedPool, pagelength, 0);
2465 MmAllocateContiguousMemorySpecifyCache(size, lowest, highest,
2466 boundary, cachetype)
2471 enum nt_caching_type cachetype;
2473 vm_memattr_t memattr;
2476 switch (cachetype) {
2478 memattr = VM_MEMATTR_UNCACHEABLE;
2480 case MmWriteCombined:
2481 memattr = VM_MEMATTR_WRITE_COMBINING;
2483 case MmNonCachedUnordered:
2484 memattr = VM_MEMATTR_UNCACHEABLE;
2487 case MmHardwareCoherentCached:
2490 memattr = VM_MEMATTR_DEFAULT;
2494 ret = (void *)kmem_alloc_contig(size, M_ZERO | M_NOWAIT, lowest,
2495 highest, PAGE_SIZE, boundary, memattr);
2497 malloc_type_allocated(M_DEVBUF, round_page(size));
2502 MmFreeContiguousMemory(base)
2509 MmFreeContiguousMemorySpecifyCache(base, size, cachetype)
2512 enum nt_caching_type cachetype;
2514 contigfree(base, size, M_DEVBUF);
2518 MmSizeOfMdl(vaddr, len)
2524 l = sizeof(struct mdl) +
2525 (sizeof(vm_offset_t *) * SPAN_PAGES(vaddr, len));
2531 * The Microsoft documentation says this routine fills in the
2532 * page array of an MDL with the _physical_ page addresses that
2533 * comprise the buffer, but we don't really want to do that here.
2534 * Instead, we just fill in the page array with the kernel virtual
2535 * addresses of the buffers.
2538 MmBuildMdlForNonPagedPool(m)
2541 vm_offset_t *mdl_pages;
2544 pagecnt = SPAN_PAGES(m->mdl_byteoffset, m->mdl_bytecount);
2546 if (pagecnt > (m->mdl_size - sizeof(mdl)) / sizeof(vm_offset_t *))
2547 panic("not enough pages in MDL to describe buffer");
2549 mdl_pages = MmGetMdlPfnArray(m);
2551 for (i = 0; i < pagecnt; i++)
2552 *mdl_pages = (vm_offset_t)m->mdl_startva + (i * PAGE_SIZE);
2554 m->mdl_flags |= MDL_SOURCE_IS_NONPAGED_POOL;
2555 m->mdl_mappedsystemva = MmGetMdlVirtualAddress(m);
2559 MmMapLockedPages(mdl *buf, uint8_t accessmode)
2561 buf->mdl_flags |= MDL_MAPPED_TO_SYSTEM_VA;
2562 return (MmGetMdlVirtualAddress(buf));
2566 MmMapLockedPagesSpecifyCache(mdl *buf, uint8_t accessmode, uint32_t cachetype,
2567 void *vaddr, uint32_t bugcheck, uint32_t prio)
2569 return (MmMapLockedPages(buf, accessmode));
2573 MmUnmapLockedPages(vaddr, buf)
2577 buf->mdl_flags &= ~MDL_MAPPED_TO_SYSTEM_VA;
2581 * This function has a problem in that it will break if you
2582 * compile this module without PAE and try to use it on a PAE
2583 * kernel. Unfortunately, there's no way around this at the
2584 * moment. It's slightly less broken that using pmap_kextract().
2585 * You'd think the virtual memory subsystem would help us out
2586 * here, but it doesn't.
2590 MmGetPhysicalAddress(void *base)
2592 return (pmap_extract(kernel_map->pmap, (vm_offset_t)base));
2596 MmGetSystemRoutineAddress(ustr)
2597 unicode_string *ustr;
2601 if (RtlUnicodeStringToAnsiString(&astr, ustr, TRUE))
2603 return (ndis_get_routine_address(ntoskrnl_functbl, astr.as_buf));
2607 MmIsAddressValid(vaddr)
2610 if (pmap_extract(kernel_map->pmap, (vm_offset_t)vaddr))
2617 MmMapIoSpace(paddr, len, cachetype)
2622 devclass_t nexus_class;
2623 device_t *nexus_devs, devp;
2624 int nexus_count = 0;
2625 device_t matching_dev = NULL;
2626 struct resource *res;
2630 /* There will always be at least one nexus. */
2632 nexus_class = devclass_find("nexus");
2633 devclass_get_devices(nexus_class, &nexus_devs, &nexus_count);
2635 for (i = 0; i < nexus_count; i++) {
2636 devp = nexus_devs[i];
2637 matching_dev = ntoskrnl_finddev(devp, paddr, &res);
2642 free(nexus_devs, M_TEMP);
2644 if (matching_dev == NULL)
2647 v = (vm_offset_t)rman_get_virtual(res);
2648 if (paddr > rman_get_start(res))
2649 v += paddr - rman_get_start(res);
2655 MmUnmapIoSpace(vaddr, len)
2663 ntoskrnl_finddev(dev, paddr, res)
2666 struct resource **res;
2668 device_t *children = NULL;
2669 device_t matching_dev;
2672 struct resource_list *rl;
2673 struct resource_list_entry *rle;
2677 /* We only want devices that have been successfully probed. */
2679 if (device_is_alive(dev) == FALSE)
2682 rl = BUS_GET_RESOURCE_LIST(device_get_parent(dev), dev);
2684 STAILQ_FOREACH(rle, rl, link) {
2690 flags = rman_get_flags(r);
2692 if (rle->type == SYS_RES_MEMORY &&
2693 paddr >= rman_get_start(r) &&
2694 paddr <= rman_get_end(r)) {
2695 if (!(flags & RF_ACTIVE))
2696 bus_activate_resource(dev,
2697 SYS_RES_MEMORY, 0, r);
2705 * If this device has children, do another
2706 * level of recursion to inspect them.
2709 device_get_children(dev, &children, &childcnt);
2711 for (i = 0; i < childcnt; i++) {
2712 matching_dev = ntoskrnl_finddev(children[i], paddr, res);
2713 if (matching_dev != NULL) {
2714 free(children, M_TEMP);
2715 return (matching_dev);
2720 /* Won't somebody please think of the children! */
2722 if (children != NULL)
2723 free(children, M_TEMP);
2729 * Workitems are unlike DPCs, in that they run in a user-mode thread
2730 * context rather than at DISPATCH_LEVEL in kernel context. In our
2731 * case we run them in kernel context anyway.
2734 ntoskrnl_workitem_thread(arg)
2744 InitializeListHead(&kq->kq_disp);
2745 kq->kq_td = curthread;
2747 KeInitializeSpinLock(&kq->kq_lock);
2748 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
2751 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
2753 KeAcquireSpinLock(&kq->kq_lock, &irql);
2757 KeReleaseSpinLock(&kq->kq_lock, irql);
2761 while (!IsListEmpty(&kq->kq_disp)) {
2762 l = RemoveHeadList(&kq->kq_disp);
2763 iw = CONTAINING_RECORD(l,
2764 io_workitem, iw_listentry);
2765 InitializeListHead((&iw->iw_listentry));
2766 if (iw->iw_func == NULL)
2768 KeReleaseSpinLock(&kq->kq_lock, irql);
2769 MSCALL2(iw->iw_func, iw->iw_dobj, iw->iw_ctx);
2770 KeAcquireSpinLock(&kq->kq_lock, &irql);
2773 KeReleaseSpinLock(&kq->kq_lock, irql);
2777 return; /* notreached */
2781 RtlCharToInteger(src, base, val)
2790 return (STATUS_ACCESS_VIOLATION);
2791 while (*src != '\0' && *src <= ' ')
2795 else if (*src == '-') {
2806 } else if (*src == 'o') {
2809 } else if (*src == 'x') {
2814 } else if (!(base == 2 || base == 8 || base == 10 || base == 16))
2815 return (STATUS_INVALID_PARAMETER);
2817 for (res = 0; *src; src++) {
2821 else if (isxdigit(*src))
2822 v = tolower(*src) - 'a' + 10;
2826 return (STATUS_INVALID_PARAMETER);
2827 res = res * base + v;
2829 *val = negative ? -res : res;
2830 return (STATUS_SUCCESS);
2834 ntoskrnl_destroy_workitem_threads(void)
2839 for (i = 0; i < WORKITEM_THREADS; i++) {
2842 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2844 tsleep(kq->kq_td->td_proc, PWAIT, "waitiw", hz/10);
2849 IoAllocateWorkItem(dobj)
2850 device_object *dobj;
2854 iw = uma_zalloc(iw_zone, M_NOWAIT);
2858 InitializeListHead(&iw->iw_listentry);
2861 mtx_lock(&ntoskrnl_dispatchlock);
2862 iw->iw_idx = wq_idx;
2863 WORKIDX_INC(wq_idx);
2864 mtx_unlock(&ntoskrnl_dispatchlock);
2873 uma_zfree(iw_zone, iw);
2877 IoQueueWorkItem(iw, iw_func, qtype, ctx)
2879 io_workitem_func iw_func;
2888 kq = wq_queues + iw->iw_idx;
2890 KeAcquireSpinLock(&kq->kq_lock, &irql);
2893 * Traverse the list and make sure this workitem hasn't
2894 * already been inserted. Queuing the same workitem
2895 * twice will hose the list but good.
2898 l = kq->kq_disp.nle_flink;
2899 while (l != &kq->kq_disp) {
2900 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2902 /* Already queued -- do nothing. */
2903 KeReleaseSpinLock(&kq->kq_lock, irql);
2909 iw->iw_func = iw_func;
2912 InsertTailList((&kq->kq_disp), (&iw->iw_listentry));
2913 KeReleaseSpinLock(&kq->kq_lock, irql);
2915 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2919 ntoskrnl_workitem(dobj, arg)
2920 device_object *dobj;
2928 w = (work_queue_item *)dobj;
2929 f = (work_item_func)w->wqi_func;
2930 uma_zfree(iw_zone, iw);
2931 MSCALL2(f, w, w->wqi_ctx);
2935 * The ExQueueWorkItem() API is deprecated in Windows XP. Microsoft
2936 * warns that it's unsafe and to use IoQueueWorkItem() instead. The
2937 * problem with ExQueueWorkItem() is that it can't guard against
2938 * the condition where a driver submits a job to the work queue and
2939 * is then unloaded before the job is able to run. IoQueueWorkItem()
2940 * acquires a reference to the device's device_object via the
2941 * object manager and retains it until after the job has completed,
2942 * which prevents the driver from being unloaded before the job
2943 * runs. (We don't currently support this behavior, though hopefully
2944 * that will change once the object manager API is fleshed out a bit.)
2946 * Having said all that, the ExQueueWorkItem() API remains, because
2947 * there are still other parts of Windows that use it, including
2948 * NDIS itself: NdisScheduleWorkItem() calls ExQueueWorkItem().
2949 * We fake up the ExQueueWorkItem() API on top of our implementation
2950 * of IoQueueWorkItem(). Workitem thread #3 is reserved exclusively
2951 * for ExQueueWorkItem() jobs, and we pass a pointer to the work
2952 * queue item (provided by the caller) in to IoAllocateWorkItem()
2953 * instead of the device_object. We need to save this pointer so
2954 * we can apply a sanity check: as with the DPC queue and other
2955 * workitem queues, we can't allow the same work queue item to
2956 * be queued twice. If it's already pending, we silently return
2960 ExQueueWorkItem(w, qtype)
2965 io_workitem_func iwf;
2973 * We need to do a special sanity test to make sure
2974 * the ExQueueWorkItem() API isn't used to queue
2975 * the same workitem twice. Rather than checking the
2976 * io_workitem pointer itself, we test the attached
2977 * device object, which is really a pointer to the
2978 * legacy work queue item structure.
2981 kq = wq_queues + WORKITEM_LEGACY_THREAD;
2982 KeAcquireSpinLock(&kq->kq_lock, &irql);
2983 l = kq->kq_disp.nle_flink;
2984 while (l != &kq->kq_disp) {
2985 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2986 if (cur->iw_dobj == (device_object *)w) {
2987 /* Already queued -- do nothing. */
2988 KeReleaseSpinLock(&kq->kq_lock, irql);
2993 KeReleaseSpinLock(&kq->kq_lock, irql);
2995 iw = IoAllocateWorkItem((device_object *)w);
2999 iw->iw_idx = WORKITEM_LEGACY_THREAD;
3000 iwf = (io_workitem_func)ntoskrnl_findwrap((funcptr)ntoskrnl_workitem);
3001 IoQueueWorkItem(iw, iwf, qtype, iw);
3005 RtlZeroMemory(dst, len)
3013 RtlSecureZeroMemory(dst, len)
3017 memset(dst, 0, len);
3021 RtlFillMemory(void *dst, size_t len, uint8_t c)
3023 memset(dst, c, len);
3027 RtlMoveMemory(dst, src, len)
3032 memmove(dst, src, len);
3036 RtlCopyMemory(dst, src, len)
3041 bcopy(src, dst, len);
3045 RtlCompareMemory(s1, s2, len)
3053 m1 = __DECONST(char *, s1);
3054 m2 = __DECONST(char *, s2);
3056 for (i = 0; i < len && m1[i] == m2[i]; i++);
3061 RtlInitAnsiString(dst, src)
3071 a->as_len = a->as_maxlen = 0;
3075 a->as_len = a->as_maxlen = strlen(src);
3080 RtlInitUnicodeString(dst, src)
3081 unicode_string *dst;
3091 u->us_len = u->us_maxlen = 0;
3098 u->us_len = u->us_maxlen = i * 2;
3103 RtlUnicodeStringToInteger(ustr, base, val)
3104 unicode_string *ustr;
3113 uchr = ustr->us_buf;
3115 bzero(abuf, sizeof(abuf));
3117 if ((char)((*uchr) & 0xFF) == '-') {
3121 } else if ((char)((*uchr) & 0xFF) == '+') {
3128 if ((char)((*uchr) & 0xFF) == 'b') {
3132 } else if ((char)((*uchr) & 0xFF) == 'o') {
3136 } else if ((char)((*uchr) & 0xFF) == 'x') {
3150 ntoskrnl_unicode_to_ascii(uchr, astr, len);
3151 *val = strtoul(abuf, NULL, base);
3153 return (STATUS_SUCCESS);
3157 RtlFreeUnicodeString(ustr)
3158 unicode_string *ustr;
3160 if (ustr->us_buf == NULL)
3162 ExFreePool(ustr->us_buf);
3163 ustr->us_buf = NULL;
3167 RtlFreeAnsiString(astr)
3170 if (astr->as_buf == NULL)
3172 ExFreePool(astr->as_buf);
3173 astr->as_buf = NULL;
3180 return (int)strtol(str, (char **)NULL, 10);
3187 return strtol(str, (char **)NULL, 10);
3198 srand(unsigned int seed)
3205 IoIsWdmVersionAvailable(uint8_t major, uint8_t minor)
3207 if (major == WDM_MAJOR && minor == WDM_MINOR_WINXP)
3213 IoOpenDeviceRegistryKey(struct device_object *devobj, uint32_t type,
3214 uint32_t mask, void **key)
3216 return (NDIS_STATUS_INVALID_DEVICE_REQUEST);
3220 IoGetDeviceObjectPointer(name, reqaccess, fileobj, devobj)
3221 unicode_string *name;
3224 device_object *devobj;
3226 return (STATUS_SUCCESS);
3230 IoGetDeviceProperty(devobj, regprop, buflen, prop, reslen)
3231 device_object *devobj;
3240 drv = devobj->do_drvobj;
3243 case DEVPROP_DRIVER_KEYNAME:
3245 *name = drv->dro_drivername.us_buf;
3246 *reslen = drv->dro_drivername.us_len;
3249 return (STATUS_INVALID_PARAMETER_2);
3253 return (STATUS_SUCCESS);
3257 KeInitializeMutex(kmutex, level)
3261 InitializeListHead((&kmutex->km_header.dh_waitlisthead));
3262 kmutex->km_abandoned = FALSE;
3263 kmutex->km_apcdisable = 1;
3264 kmutex->km_header.dh_sigstate = 1;
3265 kmutex->km_header.dh_type = DISP_TYPE_MUTANT;
3266 kmutex->km_header.dh_size = sizeof(kmutant) / sizeof(uint32_t);
3267 kmutex->km_ownerthread = NULL;
3271 KeReleaseMutex(kmutant *kmutex, uint8_t kwait)
3275 mtx_lock(&ntoskrnl_dispatchlock);
3276 prevstate = kmutex->km_header.dh_sigstate;
3277 if (kmutex->km_ownerthread != curthread) {
3278 mtx_unlock(&ntoskrnl_dispatchlock);
3279 return (STATUS_MUTANT_NOT_OWNED);
3282 kmutex->km_header.dh_sigstate++;
3283 kmutex->km_abandoned = FALSE;
3285 if (kmutex->km_header.dh_sigstate == 1) {
3286 kmutex->km_ownerthread = NULL;
3287 ntoskrnl_waittest(&kmutex->km_header, IO_NO_INCREMENT);
3290 mtx_unlock(&ntoskrnl_dispatchlock);
3296 KeReadStateMutex(kmutex)
3299 return (kmutex->km_header.dh_sigstate);
3303 KeInitializeEvent(nt_kevent *kevent, uint32_t type, uint8_t state)
3305 InitializeListHead((&kevent->k_header.dh_waitlisthead));
3306 kevent->k_header.dh_sigstate = state;
3307 if (type == EVENT_TYPE_NOTIFY)
3308 kevent->k_header.dh_type = DISP_TYPE_NOTIFICATION_EVENT;
3310 kevent->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_EVENT;
3311 kevent->k_header.dh_size = sizeof(nt_kevent) / sizeof(uint32_t);
3315 KeResetEvent(kevent)
3320 mtx_lock(&ntoskrnl_dispatchlock);
3321 prevstate = kevent->k_header.dh_sigstate;
3322 kevent->k_header.dh_sigstate = FALSE;
3323 mtx_unlock(&ntoskrnl_dispatchlock);
3329 KeSetEvent(nt_kevent *kevent, uint32_t increment, uint8_t kwait)
3333 nt_dispatch_header *dh;
3337 mtx_lock(&ntoskrnl_dispatchlock);
3338 prevstate = kevent->k_header.dh_sigstate;
3339 dh = &kevent->k_header;
3341 if (IsListEmpty(&dh->dh_waitlisthead))
3343 * If there's nobody in the waitlist, just set
3344 * the state to signalled.
3346 dh->dh_sigstate = 1;
3349 * Get the first waiter. If this is a synchronization
3350 * event, just wake up that one thread (don't bother
3351 * setting the state to signalled since we're supposed
3352 * to automatically clear synchronization events anyway).
3354 * If it's a notification event, or the first
3355 * waiter is doing a WAITTYPE_ALL wait, go through
3356 * the full wait satisfaction process.
3358 w = CONTAINING_RECORD(dh->dh_waitlisthead.nle_flink,
3359 wait_block, wb_waitlist);
3362 if (kevent->k_header.dh_type == DISP_TYPE_NOTIFICATION_EVENT ||
3363 w->wb_waittype == WAITTYPE_ALL) {
3364 if (prevstate == 0) {
3365 dh->dh_sigstate = 1;
3366 ntoskrnl_waittest(dh, increment);
3369 w->wb_awakened |= TRUE;
3370 cv_broadcastpri(&we->we_cv,
3371 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
3372 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
3376 mtx_unlock(&ntoskrnl_dispatchlock);
3382 KeClearEvent(kevent)
3385 kevent->k_header.dh_sigstate = FALSE;
3389 KeReadStateEvent(kevent)
3392 return (kevent->k_header.dh_sigstate);
3396 * The object manager in Windows is responsible for managing
3397 * references and access to various types of objects, including
3398 * device_objects, events, threads, timers and so on. However,
3399 * there's a difference in the way objects are handled in user
3400 * mode versus kernel mode.
3402 * In user mode (i.e. Win32 applications), all objects are
3403 * managed by the object manager. For example, when you create
3404 * a timer or event object, you actually end up with an
3405 * object_header (for the object manager's bookkeeping
3406 * purposes) and an object body (which contains the actual object
3407 * structure, e.g. ktimer, kevent, etc...). This allows Windows
3408 * to manage resource quotas and to enforce access restrictions
3409 * on basically every kind of system object handled by the kernel.
3411 * However, in kernel mode, you only end up using the object
3412 * manager some of the time. For example, in a driver, you create
3413 * a timer object by simply allocating the memory for a ktimer
3414 * structure and initializing it with KeInitializeTimer(). Hence,
3415 * the timer has no object_header and no reference counting or
3416 * security/resource checks are done on it. The assumption in
3417 * this case is that if you're running in kernel mode, you know
3418 * what you're doing, and you're already at an elevated privilege
3421 * There are some exceptions to this. The two most important ones
3422 * for our purposes are device_objects and threads. We need to use
3423 * the object manager to do reference counting on device_objects,
3424 * and for threads, you can only get a pointer to a thread's
3425 * dispatch header by using ObReferenceObjectByHandle() on the
3426 * handle returned by PsCreateSystemThread().
3430 ObReferenceObjectByHandle(ndis_handle handle, uint32_t reqaccess, void *otype,
3431 uint8_t accessmode, void **object, void **handleinfo)
3435 nr = malloc(sizeof(nt_objref), M_DEVBUF, M_NOWAIT|M_ZERO);
3437 return (STATUS_INSUFFICIENT_RESOURCES);
3439 InitializeListHead((&nr->no_dh.dh_waitlisthead));
3440 nr->no_obj = handle;
3441 nr->no_dh.dh_type = DISP_TYPE_THREAD;
3442 nr->no_dh.dh_sigstate = 0;
3443 nr->no_dh.dh_size = (uint8_t)(sizeof(struct thread) /
3445 TAILQ_INSERT_TAIL(&ntoskrnl_reflist, nr, link);
3448 return (STATUS_SUCCESS);
3452 ObfDereferenceObject(object)
3458 TAILQ_REMOVE(&ntoskrnl_reflist, nr, link);
3466 return (STATUS_SUCCESS);
3470 WmiQueryTraceInformation(traceclass, traceinfo, infolen, reqlen, buf)
3471 uint32_t traceclass;
3477 return (STATUS_NOT_FOUND);
3481 WmiTraceMessage(uint64_t loghandle, uint32_t messageflags,
3482 void *guid, uint16_t messagenum, ...)
3484 return (STATUS_SUCCESS);
3488 IoWMIRegistrationControl(dobj, action)
3489 device_object *dobj;
3492 return (STATUS_SUCCESS);
3496 * This is here just in case the thread returns without calling
3497 * PsTerminateSystemThread().
3500 ntoskrnl_thrfunc(arg)
3503 thread_context *thrctx;
3504 uint32_t (*tfunc)(void *);
3509 tfunc = thrctx->tc_thrfunc;
3510 tctx = thrctx->tc_thrctx;
3511 free(thrctx, M_TEMP);
3513 rval = MSCALL1(tfunc, tctx);
3515 PsTerminateSystemThread(rval);
3516 return; /* notreached */
3520 PsCreateSystemThread(handle, reqaccess, objattrs, phandle,
3521 clientid, thrfunc, thrctx)
3522 ndis_handle *handle;
3525 ndis_handle phandle;
3534 tc = malloc(sizeof(thread_context), M_TEMP, M_NOWAIT);
3536 return (STATUS_INSUFFICIENT_RESOURCES);
3538 tc->tc_thrctx = thrctx;
3539 tc->tc_thrfunc = thrfunc;
3541 error = kproc_create(ntoskrnl_thrfunc, tc, &p,
3542 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Kthread %d", ntoskrnl_kth);
3546 return (STATUS_INSUFFICIENT_RESOURCES);
3552 return (STATUS_SUCCESS);
3556 * In Windows, the exit of a thread is an event that you're allowed
3557 * to wait on, assuming you've obtained a reference to the thread using
3558 * ObReferenceObjectByHandle(). Unfortunately, the only way we can
3559 * simulate this behavior is to register each thread we create in a
3560 * reference list, and if someone holds a reference to us, we poke
3564 PsTerminateSystemThread(status)
3567 struct nt_objref *nr;
3569 mtx_lock(&ntoskrnl_dispatchlock);
3570 TAILQ_FOREACH(nr, &ntoskrnl_reflist, link) {
3571 if (nr->no_obj != curthread->td_proc)
3573 nr->no_dh.dh_sigstate = 1;
3574 ntoskrnl_waittest(&nr->no_dh, IO_NO_INCREMENT);
3577 mtx_unlock(&ntoskrnl_dispatchlock);
3582 return (0); /* notreached */
3586 DbgPrint(char *fmt, ...)
3596 return (STATUS_SUCCESS);
3603 kdb_enter(KDB_WHY_NDIS, "DbgBreakPoint(): breakpoint");
3607 KeBugCheckEx(code, param1, param2, param3, param4)
3614 panic("KeBugCheckEx: STOP 0x%X", code);
3618 ntoskrnl_timercall(arg)
3625 mtx_lock(&ntoskrnl_dispatchlock);
3629 #ifdef NTOSKRNL_DEBUG_TIMERS
3630 ntoskrnl_timer_fires++;
3632 ntoskrnl_remove_timer(timer);
3635 * This should never happen, but complain
3639 if (timer->k_header.dh_inserted == FALSE) {
3640 mtx_unlock(&ntoskrnl_dispatchlock);
3641 printf("NTOS: timer %p fired even though "
3642 "it was canceled\n", timer);
3646 /* Mark the timer as no longer being on the timer queue. */
3648 timer->k_header.dh_inserted = FALSE;
3650 /* Now signal the object and satisfy any waits on it. */
3652 timer->k_header.dh_sigstate = 1;
3653 ntoskrnl_waittest(&timer->k_header, IO_NO_INCREMENT);
3656 * If this is a periodic timer, re-arm it
3657 * so it will fire again. We do this before
3658 * calling any deferred procedure calls because
3659 * it's possible the DPC might cancel the timer,
3660 * in which case it would be wrong for us to
3661 * re-arm it again afterwards.
3664 if (timer->k_period) {
3666 tv.tv_usec = timer->k_period * 1000;
3667 timer->k_header.dh_inserted = TRUE;
3668 ntoskrnl_insert_timer(timer, tvtohz(&tv));
3669 #ifdef NTOSKRNL_DEBUG_TIMERS
3670 ntoskrnl_timer_reloads++;
3676 mtx_unlock(&ntoskrnl_dispatchlock);
3678 /* If there's a DPC associated with the timer, queue it up. */
3681 KeInsertQueueDpc(dpc, NULL, NULL);
3684 #ifdef NTOSKRNL_DEBUG_TIMERS
3686 sysctl_show_timers(SYSCTL_HANDLER_ARGS)
3691 ntoskrnl_show_timers();
3692 return (sysctl_handle_int(oidp, &ret, 0, req));
3696 ntoskrnl_show_timers()
3701 mtx_lock_spin(&ntoskrnl_calllock);
3702 l = ntoskrnl_calllist.nle_flink;
3703 while(l != &ntoskrnl_calllist) {
3707 mtx_unlock_spin(&ntoskrnl_calllock);
3710 printf("%d timers available (out of %d)\n", i, NTOSKRNL_TIMEOUTS);
3711 printf("timer sets: %qu\n", ntoskrnl_timer_sets);
3712 printf("timer reloads: %qu\n", ntoskrnl_timer_reloads);
3713 printf("timer cancels: %qu\n", ntoskrnl_timer_cancels);
3714 printf("timer fires: %qu\n", ntoskrnl_timer_fires);
3720 * Must be called with dispatcher lock held.
3724 ntoskrnl_insert_timer(timer, ticks)
3733 * Try and allocate a timer.
3735 mtx_lock_spin(&ntoskrnl_calllock);
3736 if (IsListEmpty(&ntoskrnl_calllist)) {
3737 mtx_unlock_spin(&ntoskrnl_calllock);
3738 #ifdef NTOSKRNL_DEBUG_TIMERS
3739 ntoskrnl_show_timers();
3741 panic("out of timers!");
3743 l = RemoveHeadList(&ntoskrnl_calllist);
3744 mtx_unlock_spin(&ntoskrnl_calllock);
3746 e = CONTAINING_RECORD(l, callout_entry, ce_list);
3749 timer->k_callout = c;
3752 callout_reset(c, ticks, ntoskrnl_timercall, timer);
3756 ntoskrnl_remove_timer(timer)
3761 e = (callout_entry *)timer->k_callout;
3762 callout_stop(timer->k_callout);
3764 mtx_lock_spin(&ntoskrnl_calllock);
3765 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
3766 mtx_unlock_spin(&ntoskrnl_calllock);
3770 KeInitializeTimer(timer)
3776 KeInitializeTimerEx(timer, EVENT_TYPE_NOTIFY);
3780 KeInitializeTimerEx(timer, type)
3787 bzero((char *)timer, sizeof(ktimer));
3788 InitializeListHead((&timer->k_header.dh_waitlisthead));
3789 timer->k_header.dh_sigstate = FALSE;
3790 timer->k_header.dh_inserted = FALSE;
3791 if (type == EVENT_TYPE_NOTIFY)
3792 timer->k_header.dh_type = DISP_TYPE_NOTIFICATION_TIMER;
3794 timer->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_TIMER;
3795 timer->k_header.dh_size = sizeof(ktimer) / sizeof(uint32_t);
3799 * DPC subsystem. A Windows Defered Procedure Call has the following
3801 * - It runs at DISPATCH_LEVEL.
3802 * - It can have one of 3 importance values that control when it
3803 * runs relative to other DPCs in the queue.
3804 * - On SMP systems, it can be set to run on a specific processor.
3805 * In order to satisfy the last property, we create a DPC thread for
3806 * each CPU in the system and bind it to that CPU. Each thread
3807 * maintains three queues with different importance levels, which
3808 * will be processed in order from lowest to highest.
3810 * In Windows, interrupt handlers run as DPCs. (Not to be confused
3811 * with ISRs, which run in interrupt context and can preempt DPCs.)
3812 * ISRs are given the highest importance so that they'll take
3813 * precedence over timers and other things.
3817 ntoskrnl_dpc_thread(arg)
3827 InitializeListHead(&kq->kq_disp);
3828 kq->kq_td = curthread;
3830 kq->kq_running = FALSE;
3831 KeInitializeSpinLock(&kq->kq_lock);
3832 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
3833 KeInitializeEvent(&kq->kq_done, EVENT_TYPE_SYNC, FALSE);
3836 * Elevate our priority. DPCs are used to run interrupt
3837 * handlers, and they should trigger as soon as possible
3838 * once scheduled by an ISR.
3841 thread_lock(curthread);
3842 #ifdef NTOSKRNL_MULTIPLE_DPCS
3843 sched_bind(curthread, kq->kq_cpu);
3845 sched_prio(curthread, PRI_MIN_KERN);
3846 thread_unlock(curthread);
3849 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
3851 KeAcquireSpinLock(&kq->kq_lock, &irql);
3855 KeReleaseSpinLock(&kq->kq_lock, irql);
3859 kq->kq_running = TRUE;
3861 while (!IsListEmpty(&kq->kq_disp)) {
3862 l = RemoveHeadList((&kq->kq_disp));
3863 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3864 InitializeListHead((&d->k_dpclistentry));
3865 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
3866 MSCALL4(d->k_deferedfunc, d, d->k_deferredctx,
3867 d->k_sysarg1, d->k_sysarg2);
3868 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3871 kq->kq_running = FALSE;
3873 KeReleaseSpinLock(&kq->kq_lock, irql);
3875 KeSetEvent(&kq->kq_done, IO_NO_INCREMENT, FALSE);
3879 return; /* notreached */
3883 ntoskrnl_destroy_dpc_threads(void)
3890 #ifdef NTOSKRNL_MULTIPLE_DPCS
3891 for (i = 0; i < mp_ncpus; i++) {
3893 for (i = 0; i < 1; i++) {
3898 KeInitializeDpc(&dpc, NULL, NULL);
3899 KeSetTargetProcessorDpc(&dpc, i);
3900 KeInsertQueueDpc(&dpc, NULL, NULL);
3902 tsleep(kq->kq_td->td_proc, PWAIT, "dpcw", hz/10);
3907 ntoskrnl_insert_dpc(head, dpc)
3914 l = head->nle_flink;
3916 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3922 if (dpc->k_importance == KDPC_IMPORTANCE_LOW)
3923 InsertTailList((head), (&dpc->k_dpclistentry));
3925 InsertHeadList((head), (&dpc->k_dpclistentry));
3931 KeInitializeDpc(dpc, dpcfunc, dpcctx)
3940 dpc->k_deferedfunc = dpcfunc;
3941 dpc->k_deferredctx = dpcctx;
3942 dpc->k_num = KDPC_CPU_DEFAULT;
3943 dpc->k_importance = KDPC_IMPORTANCE_MEDIUM;
3944 InitializeListHead((&dpc->k_dpclistentry));
3948 KeInsertQueueDpc(dpc, sysarg1, sysarg2)
3962 #ifdef NTOSKRNL_MULTIPLE_DPCS
3963 KeRaiseIrql(DISPATCH_LEVEL, &irql);
3966 * By default, the DPC is queued to run on the same CPU
3967 * that scheduled it.
3970 if (dpc->k_num == KDPC_CPU_DEFAULT)
3971 kq += curthread->td_oncpu;
3974 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3976 KeAcquireSpinLock(&kq->kq_lock, &irql);
3979 r = ntoskrnl_insert_dpc(&kq->kq_disp, dpc);
3981 dpc->k_sysarg1 = sysarg1;
3982 dpc->k_sysarg2 = sysarg2;
3984 KeReleaseSpinLock(&kq->kq_lock, irql);
3989 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
3995 KeRemoveQueueDpc(dpc)
4004 #ifdef NTOSKRNL_MULTIPLE_DPCS
4005 KeRaiseIrql(DISPATCH_LEVEL, &irql);
4007 kq = kq_queues + dpc->k_num;
4009 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
4012 KeAcquireSpinLock(&kq->kq_lock, &irql);
4015 if (dpc->k_dpclistentry.nle_flink == &dpc->k_dpclistentry) {
4016 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
4021 RemoveEntryList((&dpc->k_dpclistentry));
4022 InitializeListHead((&dpc->k_dpclistentry));
4024 KeReleaseSpinLock(&kq->kq_lock, irql);
4030 KeSetImportanceDpc(dpc, imp)
4034 if (imp != KDPC_IMPORTANCE_LOW &&
4035 imp != KDPC_IMPORTANCE_MEDIUM &&
4036 imp != KDPC_IMPORTANCE_HIGH)
4039 dpc->k_importance = (uint8_t)imp;
4043 KeSetTargetProcessorDpc(kdpc *dpc, uint8_t cpu)
4052 KeFlushQueuedDpcs(void)
4058 * Poke each DPC queue and wait
4059 * for them to drain.
4062 #ifdef NTOSKRNL_MULTIPLE_DPCS
4063 for (i = 0; i < mp_ncpus; i++) {
4065 for (i = 0; i < 1; i++) {
4068 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
4069 KeWaitForSingleObject(&kq->kq_done, 0, 0, TRUE, NULL);
4074 KeGetCurrentProcessorNumber(void)
4076 return ((uint32_t)curthread->td_oncpu);
4080 KeSetTimerEx(timer, duetime, period, dpc)
4093 mtx_lock(&ntoskrnl_dispatchlock);
4095 if (timer->k_header.dh_inserted == TRUE) {
4096 ntoskrnl_remove_timer(timer);
4097 #ifdef NTOSKRNL_DEBUG_TIMERS
4098 ntoskrnl_timer_cancels++;
4100 timer->k_header.dh_inserted = FALSE;
4105 timer->k_duetime = duetime;
4106 timer->k_period = period;
4107 timer->k_header.dh_sigstate = FALSE;
4111 tv.tv_sec = - (duetime) / 10000000;
4112 tv.tv_usec = (- (duetime) / 10) -
4113 (tv.tv_sec * 1000000);
4115 ntoskrnl_time(&curtime);
4116 if (duetime < curtime)
4117 tv.tv_sec = tv.tv_usec = 0;
4119 tv.tv_sec = ((duetime) - curtime) / 10000000;
4120 tv.tv_usec = ((duetime) - curtime) / 10 -
4121 (tv.tv_sec * 1000000);
4125 timer->k_header.dh_inserted = TRUE;
4126 ntoskrnl_insert_timer(timer, tvtohz(&tv));
4127 #ifdef NTOSKRNL_DEBUG_TIMERS
4128 ntoskrnl_timer_sets++;
4131 mtx_unlock(&ntoskrnl_dispatchlock);
4137 KeSetTimer(timer, duetime, dpc)
4142 return (KeSetTimerEx(timer, duetime, 0, dpc));
4146 * The Windows DDK documentation seems to say that cancelling
4147 * a timer that has a DPC will result in the DPC also being
4148 * cancelled, but this isn't really the case.
4152 KeCancelTimer(timer)
4160 mtx_lock(&ntoskrnl_dispatchlock);
4162 pending = timer->k_header.dh_inserted;
4164 if (timer->k_header.dh_inserted == TRUE) {
4165 timer->k_header.dh_inserted = FALSE;
4166 ntoskrnl_remove_timer(timer);
4167 #ifdef NTOSKRNL_DEBUG_TIMERS
4168 ntoskrnl_timer_cancels++;
4172 mtx_unlock(&ntoskrnl_dispatchlock);
4178 KeReadStateTimer(timer)
4181 return (timer->k_header.dh_sigstate);
4185 KeDelayExecutionThread(uint8_t wait_mode, uint8_t alertable, int64_t *interval)
4190 panic("invalid wait_mode %d", wait_mode);
4192 KeInitializeTimer(&timer);
4193 KeSetTimer(&timer, *interval, NULL);
4194 KeWaitForSingleObject(&timer, 0, 0, alertable, NULL);
4196 return STATUS_SUCCESS;
4200 KeQueryInterruptTime(void)
4205 getmicrouptime(&tv);
4207 ticks = tvtohz(&tv);
4209 return ticks * howmany(10000000, hz);
4212 static struct thread *
4213 KeGetCurrentThread(void)
4220 KeSetPriorityThread(td, pri)
4227 return LOW_REALTIME_PRIORITY;
4229 if (td->td_priority <= PRI_MIN_KERN)
4230 old = HIGH_PRIORITY;
4231 else if (td->td_priority >= PRI_MAX_KERN)
4234 old = LOW_REALTIME_PRIORITY;
4237 if (pri == HIGH_PRIORITY)
4238 sched_prio(td, PRI_MIN_KERN);
4239 if (pri == LOW_REALTIME_PRIORITY)
4240 sched_prio(td, PRI_MIN_KERN + (PRI_MAX_KERN - PRI_MIN_KERN) / 2);
4241 if (pri == LOW_PRIORITY)
4242 sched_prio(td, PRI_MAX_KERN);
4251 printf("ntoskrnl dummy called...\n");
4255 image_patch_table ntoskrnl_functbl[] = {
4256 IMPORT_SFUNC(RtlZeroMemory, 2),
4257 IMPORT_SFUNC(RtlSecureZeroMemory, 2),
4258 IMPORT_SFUNC(RtlFillMemory, 3),
4259 IMPORT_SFUNC(RtlMoveMemory, 3),
4260 IMPORT_SFUNC(RtlCharToInteger, 3),
4261 IMPORT_SFUNC(RtlCopyMemory, 3),
4262 IMPORT_SFUNC(RtlCopyString, 2),
4263 IMPORT_SFUNC(RtlCompareMemory, 3),
4264 IMPORT_SFUNC(RtlEqualUnicodeString, 3),
4265 IMPORT_SFUNC(RtlCopyUnicodeString, 2),
4266 IMPORT_SFUNC(RtlUnicodeStringToAnsiString, 3),
4267 IMPORT_SFUNC(RtlAnsiStringToUnicodeString, 3),
4268 IMPORT_SFUNC(RtlInitAnsiString, 2),
4269 IMPORT_SFUNC_MAP(RtlInitString, RtlInitAnsiString, 2),
4270 IMPORT_SFUNC(RtlInitUnicodeString, 2),
4271 IMPORT_SFUNC(RtlFreeAnsiString, 1),
4272 IMPORT_SFUNC(RtlFreeUnicodeString, 1),
4273 IMPORT_SFUNC(RtlUnicodeStringToInteger, 3),
4274 IMPORT_CFUNC(sprintf, 0),
4275 IMPORT_CFUNC(vsprintf, 0),
4276 IMPORT_CFUNC_MAP(_snprintf, snprintf, 0),
4277 IMPORT_CFUNC_MAP(_vsnprintf, vsnprintf, 0),
4278 IMPORT_CFUNC(DbgPrint, 0),
4279 IMPORT_SFUNC(DbgBreakPoint, 0),
4280 IMPORT_SFUNC(KeBugCheckEx, 5),
4281 IMPORT_CFUNC(strncmp, 0),
4282 IMPORT_CFUNC(strcmp, 0),
4283 IMPORT_CFUNC_MAP(stricmp, strcasecmp, 0),
4284 IMPORT_CFUNC(strncpy, 0),
4285 IMPORT_CFUNC(strcpy, 0),
4286 IMPORT_CFUNC(strlen, 0),
4287 IMPORT_CFUNC_MAP(toupper, ntoskrnl_toupper, 0),
4288 IMPORT_CFUNC_MAP(tolower, ntoskrnl_tolower, 0),
4289 IMPORT_CFUNC_MAP(strstr, ntoskrnl_strstr, 0),
4290 IMPORT_CFUNC_MAP(strncat, ntoskrnl_strncat, 0),
4291 IMPORT_CFUNC_MAP(strchr, index, 0),
4292 IMPORT_CFUNC_MAP(strrchr, rindex, 0),
4293 IMPORT_CFUNC(memcpy, 0),
4294 IMPORT_CFUNC_MAP(memmove, ntoskrnl_memmove, 0),
4295 IMPORT_CFUNC_MAP(memset, ntoskrnl_memset, 0),
4296 IMPORT_CFUNC_MAP(memchr, ntoskrnl_memchr, 0),
4297 IMPORT_SFUNC(IoAllocateDriverObjectExtension, 4),
4298 IMPORT_SFUNC(IoGetDriverObjectExtension, 2),
4299 IMPORT_FFUNC(IofCallDriver, 2),
4300 IMPORT_FFUNC(IofCompleteRequest, 2),
4301 IMPORT_SFUNC(IoAcquireCancelSpinLock, 1),
4302 IMPORT_SFUNC(IoReleaseCancelSpinLock, 1),
4303 IMPORT_SFUNC(IoCancelIrp, 1),
4304 IMPORT_SFUNC(IoConnectInterrupt, 11),
4305 IMPORT_SFUNC(IoDisconnectInterrupt, 1),
4306 IMPORT_SFUNC(IoCreateDevice, 7),
4307 IMPORT_SFUNC(IoDeleteDevice, 1),
4308 IMPORT_SFUNC(IoGetAttachedDevice, 1),
4309 IMPORT_SFUNC(IoAttachDeviceToDeviceStack, 2),
4310 IMPORT_SFUNC(IoDetachDevice, 1),
4311 IMPORT_SFUNC(IoBuildSynchronousFsdRequest, 7),
4312 IMPORT_SFUNC(IoBuildAsynchronousFsdRequest, 6),
4313 IMPORT_SFUNC(IoBuildDeviceIoControlRequest, 9),
4314 IMPORT_SFUNC(IoAllocateIrp, 2),
4315 IMPORT_SFUNC(IoReuseIrp, 2),
4316 IMPORT_SFUNC(IoMakeAssociatedIrp, 2),
4317 IMPORT_SFUNC(IoFreeIrp, 1),
4318 IMPORT_SFUNC(IoInitializeIrp, 3),
4319 IMPORT_SFUNC(KeAcquireInterruptSpinLock, 1),
4320 IMPORT_SFUNC(KeReleaseInterruptSpinLock, 2),
4321 IMPORT_SFUNC(KeSynchronizeExecution, 3),
4322 IMPORT_SFUNC(KeWaitForSingleObject, 5),
4323 IMPORT_SFUNC(KeWaitForMultipleObjects, 8),
4324 IMPORT_SFUNC(_allmul, 4),
4325 IMPORT_SFUNC(_alldiv, 4),
4326 IMPORT_SFUNC(_allrem, 4),
4327 IMPORT_RFUNC(_allshr, 0),
4328 IMPORT_RFUNC(_allshl, 0),
4329 IMPORT_SFUNC(_aullmul, 4),
4330 IMPORT_SFUNC(_aulldiv, 4),
4331 IMPORT_SFUNC(_aullrem, 4),
4332 IMPORT_RFUNC(_aullshr, 0),
4333 IMPORT_RFUNC(_aullshl, 0),
4334 IMPORT_CFUNC(atoi, 0),
4335 IMPORT_CFUNC(atol, 0),
4336 IMPORT_CFUNC(rand, 0),
4337 IMPORT_CFUNC(srand, 0),
4338 IMPORT_SFUNC(WRITE_REGISTER_USHORT, 2),
4339 IMPORT_SFUNC(READ_REGISTER_USHORT, 1),
4340 IMPORT_SFUNC(WRITE_REGISTER_ULONG, 2),
4341 IMPORT_SFUNC(READ_REGISTER_ULONG, 1),
4342 IMPORT_SFUNC(READ_REGISTER_UCHAR, 1),
4343 IMPORT_SFUNC(WRITE_REGISTER_UCHAR, 2),
4344 IMPORT_SFUNC(ExInitializePagedLookasideList, 7),
4345 IMPORT_SFUNC(ExDeletePagedLookasideList, 1),
4346 IMPORT_SFUNC(ExInitializeNPagedLookasideList, 7),
4347 IMPORT_SFUNC(ExDeleteNPagedLookasideList, 1),
4348 IMPORT_FFUNC(InterlockedPopEntrySList, 1),
4349 IMPORT_FFUNC(InitializeSListHead, 1),
4350 IMPORT_FFUNC(InterlockedPushEntrySList, 2),
4351 IMPORT_SFUNC(ExQueryDepthSList, 1),
4352 IMPORT_FFUNC_MAP(ExpInterlockedPopEntrySList,
4353 InterlockedPopEntrySList, 1),
4354 IMPORT_FFUNC_MAP(ExpInterlockedPushEntrySList,
4355 InterlockedPushEntrySList, 2),
4356 IMPORT_FFUNC(ExInterlockedPopEntrySList, 2),
4357 IMPORT_FFUNC(ExInterlockedPushEntrySList, 3),
4358 IMPORT_SFUNC(ExAllocatePoolWithTag, 3),
4359 IMPORT_SFUNC(ExFreePoolWithTag, 2),
4360 IMPORT_SFUNC(ExFreePool, 1),
4362 IMPORT_FFUNC(KefAcquireSpinLockAtDpcLevel, 1),
4363 IMPORT_FFUNC(KefReleaseSpinLockFromDpcLevel,1),
4364 IMPORT_FFUNC(KeAcquireSpinLockRaiseToDpc, 1),
4367 * For AMD64, we can get away with just mapping
4368 * KeAcquireSpinLockRaiseToDpc() directly to KfAcquireSpinLock()
4369 * because the calling conventions end up being the same.
4370 * On i386, we have to be careful because KfAcquireSpinLock()
4371 * is _fastcall but KeAcquireSpinLockRaiseToDpc() isn't.
4373 IMPORT_SFUNC(KeAcquireSpinLockAtDpcLevel, 1),
4374 IMPORT_SFUNC(KeReleaseSpinLockFromDpcLevel, 1),
4375 IMPORT_SFUNC_MAP(KeAcquireSpinLockRaiseToDpc, KfAcquireSpinLock, 1),
4377 IMPORT_SFUNC_MAP(KeReleaseSpinLock, KfReleaseSpinLock, 1),
4378 IMPORT_FFUNC(InterlockedIncrement, 1),
4379 IMPORT_FFUNC(InterlockedDecrement, 1),
4380 IMPORT_FFUNC(InterlockedExchange, 2),
4381 IMPORT_FFUNC(ExInterlockedAddLargeStatistic, 2),
4382 IMPORT_SFUNC(IoAllocateMdl, 5),
4383 IMPORT_SFUNC(IoFreeMdl, 1),
4384 IMPORT_SFUNC(MmAllocateContiguousMemory, 2 + 1),
4385 IMPORT_SFUNC(MmAllocateContiguousMemorySpecifyCache, 5 + 3),
4386 IMPORT_SFUNC(MmFreeContiguousMemory, 1),
4387 IMPORT_SFUNC(MmFreeContiguousMemorySpecifyCache, 3),
4388 IMPORT_SFUNC(MmSizeOfMdl, 1),
4389 IMPORT_SFUNC(MmMapLockedPages, 2),
4390 IMPORT_SFUNC(MmMapLockedPagesSpecifyCache, 6),
4391 IMPORT_SFUNC(MmUnmapLockedPages, 2),
4392 IMPORT_SFUNC(MmBuildMdlForNonPagedPool, 1),
4393 IMPORT_SFUNC(MmGetPhysicalAddress, 1),
4394 IMPORT_SFUNC(MmGetSystemRoutineAddress, 1),
4395 IMPORT_SFUNC(MmIsAddressValid, 1),
4396 IMPORT_SFUNC(MmMapIoSpace, 3 + 1),
4397 IMPORT_SFUNC(MmUnmapIoSpace, 2),
4398 IMPORT_SFUNC(KeInitializeSpinLock, 1),
4399 IMPORT_SFUNC(IoIsWdmVersionAvailable, 2),
4400 IMPORT_SFUNC(IoOpenDeviceRegistryKey, 4),
4401 IMPORT_SFUNC(IoGetDeviceObjectPointer, 4),
4402 IMPORT_SFUNC(IoGetDeviceProperty, 5),
4403 IMPORT_SFUNC(IoAllocateWorkItem, 1),
4404 IMPORT_SFUNC(IoFreeWorkItem, 1),
4405 IMPORT_SFUNC(IoQueueWorkItem, 4),
4406 IMPORT_SFUNC(ExQueueWorkItem, 2),
4407 IMPORT_SFUNC(ntoskrnl_workitem, 2),
4408 IMPORT_SFUNC(KeInitializeMutex, 2),
4409 IMPORT_SFUNC(KeReleaseMutex, 2),
4410 IMPORT_SFUNC(KeReadStateMutex, 1),
4411 IMPORT_SFUNC(KeInitializeEvent, 3),
4412 IMPORT_SFUNC(KeSetEvent, 3),
4413 IMPORT_SFUNC(KeResetEvent, 1),
4414 IMPORT_SFUNC(KeClearEvent, 1),
4415 IMPORT_SFUNC(KeReadStateEvent, 1),
4416 IMPORT_SFUNC(KeInitializeTimer, 1),
4417 IMPORT_SFUNC(KeInitializeTimerEx, 2),
4418 IMPORT_SFUNC(KeSetTimer, 3),
4419 IMPORT_SFUNC(KeSetTimerEx, 4),
4420 IMPORT_SFUNC(KeCancelTimer, 1),
4421 IMPORT_SFUNC(KeReadStateTimer, 1),
4422 IMPORT_SFUNC(KeInitializeDpc, 3),
4423 IMPORT_SFUNC(KeInsertQueueDpc, 3),
4424 IMPORT_SFUNC(KeRemoveQueueDpc, 1),
4425 IMPORT_SFUNC(KeSetImportanceDpc, 2),
4426 IMPORT_SFUNC(KeSetTargetProcessorDpc, 2),
4427 IMPORT_SFUNC(KeFlushQueuedDpcs, 0),
4428 IMPORT_SFUNC(KeGetCurrentProcessorNumber, 1),
4429 IMPORT_SFUNC(ObReferenceObjectByHandle, 6),
4430 IMPORT_FFUNC(ObfDereferenceObject, 1),
4431 IMPORT_SFUNC(ZwClose, 1),
4432 IMPORT_SFUNC(PsCreateSystemThread, 7),
4433 IMPORT_SFUNC(PsTerminateSystemThread, 1),
4434 IMPORT_SFUNC(IoWMIRegistrationControl, 2),
4435 IMPORT_SFUNC(WmiQueryTraceInformation, 5),
4436 IMPORT_CFUNC(WmiTraceMessage, 0),
4437 IMPORT_SFUNC(KeQuerySystemTime, 1),
4438 IMPORT_CFUNC(KeTickCount, 0),
4439 IMPORT_SFUNC(KeDelayExecutionThread, 3),
4440 IMPORT_SFUNC(KeQueryInterruptTime, 0),
4441 IMPORT_SFUNC(KeGetCurrentThread, 0),
4442 IMPORT_SFUNC(KeSetPriorityThread, 2),
4445 * This last entry is a catch-all for any function we haven't
4446 * implemented yet. The PE import list patching routine will
4447 * use it for any function that doesn't have an explicit match
4451 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
4455 { NULL, NULL, NULL }