3 * Bill Paul <wpaul@windriver.com>. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by Bill Paul.
16 * 4. Neither the name of the author nor the names of any co-contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30 * THE POSSIBILITY OF SUCH DAMAGE.
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include <sys/ctype.h>
37 #include <sys/unistd.h>
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/errno.h>
41 #include <sys/systm.h>
42 #include <sys/malloc.h>
44 #include <sys/mutex.h>
46 #include <sys/callout.h>
48 #include <sys/kernel.h>
50 #include <sys/condvar.h>
51 #include <sys/kthread.h>
52 #include <sys/module.h>
54 #include <sys/sched.h>
55 #include <sys/sysctl.h>
57 #include <machine/atomic.h>
58 #include <machine/bus.h>
59 #include <machine/stdarg.h>
60 #include <machine/resource.h>
66 #include <vm/vm_param.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_extern.h>
73 #include <compat/ndis/pe_var.h>
74 #include <compat/ndis/cfg_var.h>
75 #include <compat/ndis/resource_var.h>
76 #include <compat/ndis/ntoskrnl_var.h>
77 #include <compat/ndis/hal_var.h>
78 #include <compat/ndis/ndis_var.h>
80 #ifdef NTOSKRNL_DEBUG_TIMERS
81 static int sysctl_show_timers(SYSCTL_HANDLER_ARGS);
83 SYSCTL_PROC(_debug, OID_AUTO, ntoskrnl_timers, CTLTYPE_INT | CTLFLAG_RW,
84 NULL, 0, sysctl_show_timers, "I",
85 "Show ntoskrnl timer stats");
99 typedef struct kdpc_queue kdpc_queue;
103 struct thread *we_td;
106 typedef struct wb_ext wb_ext;
108 #define NTOSKRNL_TIMEOUTS 256
109 #ifdef NTOSKRNL_DEBUG_TIMERS
110 static uint64_t ntoskrnl_timer_fires;
111 static uint64_t ntoskrnl_timer_sets;
112 static uint64_t ntoskrnl_timer_reloads;
113 static uint64_t ntoskrnl_timer_cancels;
116 struct callout_entry {
117 struct callout ce_callout;
121 typedef struct callout_entry callout_entry;
123 static struct list_entry ntoskrnl_calllist;
124 static struct mtx ntoskrnl_calllock;
125 struct kuser_shared_data kuser_shared_data;
127 static struct list_entry ntoskrnl_intlist;
128 static kspin_lock ntoskrnl_intlock;
130 static uint8_t RtlEqualUnicodeString(unicode_string *,
131 unicode_string *, uint8_t);
132 static void RtlCopyString(ansi_string *, const ansi_string *);
133 static void RtlCopyUnicodeString(unicode_string *,
135 static irp *IoBuildSynchronousFsdRequest(uint32_t, device_object *,
136 void *, uint32_t, uint64_t *, nt_kevent *, io_status_block *);
137 static irp *IoBuildAsynchronousFsdRequest(uint32_t,
138 device_object *, void *, uint32_t, uint64_t *, io_status_block *);
139 static irp *IoBuildDeviceIoControlRequest(uint32_t,
140 device_object *, void *, uint32_t, void *, uint32_t,
141 uint8_t, nt_kevent *, io_status_block *);
142 static irp *IoAllocateIrp(uint8_t, uint8_t);
143 static void IoReuseIrp(irp *, uint32_t);
144 static void IoFreeIrp(irp *);
145 static void IoInitializeIrp(irp *, uint16_t, uint8_t);
146 static irp *IoMakeAssociatedIrp(irp *, uint8_t);
147 static uint32_t KeWaitForMultipleObjects(uint32_t,
148 nt_dispatch_header **, uint32_t, uint32_t, uint32_t, uint8_t,
149 int64_t *, wait_block *);
150 static void ntoskrnl_waittest(nt_dispatch_header *, uint32_t);
151 static void ntoskrnl_satisfy_wait(nt_dispatch_header *, struct thread *);
152 static void ntoskrnl_satisfy_multiple_waits(wait_block *);
153 static int ntoskrnl_is_signalled(nt_dispatch_header *, struct thread *);
154 static void ntoskrnl_insert_timer(ktimer *, int);
155 static void ntoskrnl_remove_timer(ktimer *);
156 #ifdef NTOSKRNL_DEBUG_TIMERS
157 static void ntoskrnl_show_timers(void);
159 static void ntoskrnl_timercall(void *);
160 static void ntoskrnl_dpc_thread(void *);
161 static void ntoskrnl_destroy_dpc_threads(void);
162 static void ntoskrnl_destroy_workitem_threads(void);
163 static void ntoskrnl_workitem_thread(void *);
164 static void ntoskrnl_workitem(device_object *, void *);
165 static void ntoskrnl_unicode_to_ascii(uint16_t *, char *, int);
166 static void ntoskrnl_ascii_to_unicode(char *, uint16_t *, int);
167 static uint8_t ntoskrnl_insert_dpc(list_entry *, kdpc *);
168 static void WRITE_REGISTER_USHORT(uint16_t *, uint16_t);
169 static uint16_t READ_REGISTER_USHORT(uint16_t *);
170 static void WRITE_REGISTER_ULONG(uint32_t *, uint32_t);
171 static uint32_t READ_REGISTER_ULONG(uint32_t *);
172 static void WRITE_REGISTER_UCHAR(uint8_t *, uint8_t);
173 static uint8_t READ_REGISTER_UCHAR(uint8_t *);
174 static int64_t _allmul(int64_t, int64_t);
175 static int64_t _alldiv(int64_t, int64_t);
176 static int64_t _allrem(int64_t, int64_t);
177 static int64_t _allshr(int64_t, uint8_t);
178 static int64_t _allshl(int64_t, uint8_t);
179 static uint64_t _aullmul(uint64_t, uint64_t);
180 static uint64_t _aulldiv(uint64_t, uint64_t);
181 static uint64_t _aullrem(uint64_t, uint64_t);
182 static uint64_t _aullshr(uint64_t, uint8_t);
183 static uint64_t _aullshl(uint64_t, uint8_t);
184 static slist_entry *ntoskrnl_pushsl(slist_header *, slist_entry *);
185 static void InitializeSListHead(slist_header *);
186 static slist_entry *ntoskrnl_popsl(slist_header *);
187 static void ExFreePoolWithTag(void *, uint32_t);
188 static void ExInitializePagedLookasideList(paged_lookaside_list *,
189 lookaside_alloc_func *, lookaside_free_func *,
190 uint32_t, size_t, uint32_t, uint16_t);
191 static void ExDeletePagedLookasideList(paged_lookaside_list *);
192 static void ExInitializeNPagedLookasideList(npaged_lookaside_list *,
193 lookaside_alloc_func *, lookaside_free_func *,
194 uint32_t, size_t, uint32_t, uint16_t);
195 static void ExDeleteNPagedLookasideList(npaged_lookaside_list *);
197 *ExInterlockedPushEntrySList(slist_header *,
198 slist_entry *, kspin_lock *);
200 *ExInterlockedPopEntrySList(slist_header *, kspin_lock *);
201 static uint32_t InterlockedIncrement(volatile uint32_t *);
202 static uint32_t InterlockedDecrement(volatile uint32_t *);
203 static void ExInterlockedAddLargeStatistic(uint64_t *, uint32_t);
204 static void *MmAllocateContiguousMemory(uint32_t, uint64_t);
205 static void *MmAllocateContiguousMemorySpecifyCache(uint32_t,
206 uint64_t, uint64_t, uint64_t, enum nt_caching_type);
207 static void MmFreeContiguousMemory(void *);
208 static void MmFreeContiguousMemorySpecifyCache(void *, uint32_t,
209 enum nt_caching_type);
210 static uint32_t MmSizeOfMdl(void *, size_t);
211 static void *MmMapLockedPages(mdl *, uint8_t);
212 static void *MmMapLockedPagesSpecifyCache(mdl *,
213 uint8_t, uint32_t, void *, uint32_t, uint32_t);
214 static void MmUnmapLockedPages(void *, mdl *);
215 static device_t ntoskrnl_finddev(device_t, uint64_t, struct resource **);
216 static void RtlZeroMemory(void *, size_t);
217 static void RtlSecureZeroMemory(void *, size_t);
218 static void RtlFillMemory(void *, size_t, uint8_t);
219 static void RtlMoveMemory(void *, const void *, size_t);
220 static ndis_status RtlCharToInteger(const char *, uint32_t, uint32_t *);
221 static void RtlCopyMemory(void *, const void *, size_t);
222 static size_t RtlCompareMemory(const void *, const void *, size_t);
223 static ndis_status RtlUnicodeStringToInteger(unicode_string *,
224 uint32_t, uint32_t *);
225 static int atoi (const char *);
226 static long atol (const char *);
227 static int rand(void);
228 static void srand(unsigned int);
229 static void KeQuerySystemTime(uint64_t *);
230 static uint32_t KeTickCount(void);
231 static uint8_t IoIsWdmVersionAvailable(uint8_t, uint8_t);
232 static int32_t IoOpenDeviceRegistryKey(struct device_object *, uint32_t,
234 static void ntoskrnl_thrfunc(void *);
235 static ndis_status PsCreateSystemThread(ndis_handle *,
236 uint32_t, void *, ndis_handle, void *, void *, void *);
237 static ndis_status PsTerminateSystemThread(ndis_status);
238 static ndis_status IoGetDeviceObjectPointer(unicode_string *,
239 uint32_t, void *, device_object *);
240 static ndis_status IoGetDeviceProperty(device_object *, uint32_t,
241 uint32_t, void *, uint32_t *);
242 static void KeInitializeMutex(kmutant *, uint32_t);
243 static uint32_t KeReleaseMutex(kmutant *, uint8_t);
244 static uint32_t KeReadStateMutex(kmutant *);
245 static ndis_status ObReferenceObjectByHandle(ndis_handle,
246 uint32_t, void *, uint8_t, void **, void **);
247 static void ObfDereferenceObject(void *);
248 static uint32_t ZwClose(ndis_handle);
249 static uint32_t WmiQueryTraceInformation(uint32_t, void *, uint32_t,
251 static uint32_t WmiTraceMessage(uint64_t, uint32_t, void *, uint16_t, ...);
252 static uint32_t IoWMIRegistrationControl(device_object *, uint32_t);
253 static void *ntoskrnl_memset(void *, int, size_t);
254 static void *ntoskrnl_memmove(void *, void *, size_t);
255 static void *ntoskrnl_memchr(void *, unsigned char, size_t);
256 static char *ntoskrnl_strstr(char *, char *);
257 static char *ntoskrnl_strncat(char *, char *, size_t);
258 static int ntoskrnl_toupper(int);
259 static int ntoskrnl_tolower(int);
260 static funcptr ntoskrnl_findwrap(funcptr);
261 static uint32_t DbgPrint(char *, ...);
262 static void DbgBreakPoint(void);
263 static void KeBugCheckEx(uint32_t, u_long, u_long, u_long, u_long);
264 static int32_t KeDelayExecutionThread(uint8_t, uint8_t, int64_t *);
265 static int32_t KeSetPriorityThread(struct thread *, int32_t);
266 static void dummy(void);
268 static struct mtx ntoskrnl_dispatchlock;
269 static struct mtx ntoskrnl_interlock;
270 static kspin_lock ntoskrnl_cancellock;
271 static int ntoskrnl_kth = 0;
272 static struct nt_objref_head ntoskrnl_reflist;
273 static uma_zone_t mdl_zone;
274 static uma_zone_t iw_zone;
275 static struct kdpc_queue *kq_queues;
276 static struct kdpc_queue *wq_queues;
277 static int wq_idx = 0;
282 image_patch_table *patch;
289 mtx_init(&ntoskrnl_dispatchlock,
290 "ntoskrnl dispatch lock", MTX_NDIS_LOCK, MTX_DEF|MTX_RECURSE);
291 mtx_init(&ntoskrnl_interlock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
292 KeInitializeSpinLock(&ntoskrnl_cancellock);
293 KeInitializeSpinLock(&ntoskrnl_intlock);
294 TAILQ_INIT(&ntoskrnl_reflist);
296 InitializeListHead(&ntoskrnl_calllist);
297 InitializeListHead(&ntoskrnl_intlist);
298 mtx_init(&ntoskrnl_calllock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
300 kq_queues = ExAllocatePoolWithTag(NonPagedPool,
301 #ifdef NTOSKRNL_MULTIPLE_DPCS
302 sizeof(kdpc_queue) * mp_ncpus, 0);
304 sizeof(kdpc_queue), 0);
307 if (kq_queues == NULL)
310 wq_queues = ExAllocatePoolWithTag(NonPagedPool,
311 sizeof(kdpc_queue) * WORKITEM_THREADS, 0);
313 if (wq_queues == NULL)
316 #ifdef NTOSKRNL_MULTIPLE_DPCS
317 bzero((char *)kq_queues, sizeof(kdpc_queue) * mp_ncpus);
319 bzero((char *)kq_queues, sizeof(kdpc_queue));
321 bzero((char *)wq_queues, sizeof(kdpc_queue) * WORKITEM_THREADS);
324 * Launch the DPC threads.
327 #ifdef NTOSKRNL_MULTIPLE_DPCS
328 for (i = 0; i < mp_ncpus; i++) {
330 for (i = 0; i < 1; i++) {
334 error = kproc_create(ntoskrnl_dpc_thread, kq, &p,
335 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows DPC %d", i);
337 panic("failed to launch DPC thread");
341 * Launch the workitem threads.
344 for (i = 0; i < WORKITEM_THREADS; i++) {
346 error = kproc_create(ntoskrnl_workitem_thread, kq, &p,
347 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Workitem %d", i);
349 panic("failed to launch workitem thread");
352 patch = ntoskrnl_functbl;
353 while (patch->ipt_func != NULL) {
354 windrv_wrap((funcptr)patch->ipt_func,
355 (funcptr *)&patch->ipt_wrap,
356 patch->ipt_argcnt, patch->ipt_ftype);
360 for (i = 0; i < NTOSKRNL_TIMEOUTS; i++) {
361 e = ExAllocatePoolWithTag(NonPagedPool,
362 sizeof(callout_entry), 0);
364 panic("failed to allocate timeouts");
365 mtx_lock_spin(&ntoskrnl_calllock);
366 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
367 mtx_unlock_spin(&ntoskrnl_calllock);
371 * MDLs are supposed to be variable size (they describe
372 * buffers containing some number of pages, but we don't
373 * know ahead of time how many pages that will be). But
374 * always allocating them off the heap is very slow. As
375 * a compromise, we create an MDL UMA zone big enough to
376 * handle any buffer requiring up to 16 pages, and we
377 * use those for any MDLs for buffers of 16 pages or less
378 * in size. For buffers larger than that (which we assume
379 * will be few and far between, we allocate the MDLs off
383 mdl_zone = uma_zcreate("Windows MDL", MDL_ZONE_SIZE,
384 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
386 iw_zone = uma_zcreate("Windows WorkItem", sizeof(io_workitem),
387 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
395 image_patch_table *patch;
399 patch = ntoskrnl_functbl;
400 while (patch->ipt_func != NULL) {
401 windrv_unwrap(patch->ipt_wrap);
405 /* Stop the workitem queues. */
406 ntoskrnl_destroy_workitem_threads();
407 /* Stop the DPC queues. */
408 ntoskrnl_destroy_dpc_threads();
410 ExFreePool(kq_queues);
411 ExFreePool(wq_queues);
413 uma_zdestroy(mdl_zone);
414 uma_zdestroy(iw_zone);
416 mtx_lock_spin(&ntoskrnl_calllock);
417 while(!IsListEmpty(&ntoskrnl_calllist)) {
418 l = RemoveHeadList(&ntoskrnl_calllist);
419 e = CONTAINING_RECORD(l, callout_entry, ce_list);
420 mtx_unlock_spin(&ntoskrnl_calllock);
422 mtx_lock_spin(&ntoskrnl_calllock);
424 mtx_unlock_spin(&ntoskrnl_calllock);
426 mtx_destroy(&ntoskrnl_dispatchlock);
427 mtx_destroy(&ntoskrnl_interlock);
428 mtx_destroy(&ntoskrnl_calllock);
434 * We need to be able to reference this externally from the wrapper;
435 * GCC only generates a local implementation of memset.
438 ntoskrnl_memset(buf, ch, size)
443 return (memset(buf, ch, size));
447 ntoskrnl_memmove(dst, src, size)
452 bcopy(src, dst, size);
457 ntoskrnl_memchr(void *buf, unsigned char ch, size_t len)
460 unsigned char *p = buf;
465 } while (--len != 0);
471 ntoskrnl_strstr(s, find)
477 if ((c = *find++) != 0) {
481 if ((sc = *s++) == 0)
484 } while (strncmp(s, find, len) != 0);
490 /* Taken from libc */
492 ntoskrnl_strncat(dst, src, n)
504 if ((*d = *s++) == 0)
528 RtlEqualUnicodeString(unicode_string *str1, unicode_string *str2,
529 uint8_t caseinsensitive)
533 if (str1->us_len != str2->us_len)
536 for (i = 0; i < str1->us_len; i++) {
537 if (caseinsensitive == TRUE) {
538 if (toupper((char)(str1->us_buf[i] & 0xFF)) !=
539 toupper((char)(str2->us_buf[i] & 0xFF)))
542 if (str1->us_buf[i] != str2->us_buf[i])
551 RtlCopyString(dst, src)
553 const ansi_string *src;
555 if (src != NULL && src->as_buf != NULL && dst->as_buf != NULL) {
556 dst->as_len = min(src->as_len, dst->as_maxlen);
557 memcpy(dst->as_buf, src->as_buf, dst->as_len);
558 if (dst->as_len < dst->as_maxlen)
559 dst->as_buf[dst->as_len] = 0;
565 RtlCopyUnicodeString(dest, src)
566 unicode_string *dest;
570 if (dest->us_maxlen >= src->us_len)
571 dest->us_len = src->us_len;
573 dest->us_len = dest->us_maxlen;
574 memcpy(dest->us_buf, src->us_buf, dest->us_len);
578 ntoskrnl_ascii_to_unicode(ascii, unicode, len)
587 for (i = 0; i < len; i++) {
588 *ustr = (uint16_t)ascii[i];
594 ntoskrnl_unicode_to_ascii(unicode, ascii, len)
603 for (i = 0; i < len / 2; i++) {
604 *astr = (uint8_t)unicode[i];
610 RtlUnicodeStringToAnsiString(ansi_string *dest, unicode_string *src, uint8_t allocate)
612 if (dest == NULL || src == NULL)
613 return (STATUS_INVALID_PARAMETER);
615 dest->as_len = src->us_len / 2;
616 if (dest->as_maxlen < dest->as_len)
617 dest->as_len = dest->as_maxlen;
619 if (allocate == TRUE) {
620 dest->as_buf = ExAllocatePoolWithTag(NonPagedPool,
621 (src->us_len / 2) + 1, 0);
622 if (dest->as_buf == NULL)
623 return (STATUS_INSUFFICIENT_RESOURCES);
624 dest->as_len = dest->as_maxlen = src->us_len / 2;
626 dest->as_len = src->us_len / 2; /* XXX */
627 if (dest->as_maxlen < dest->as_len)
628 dest->as_len = dest->as_maxlen;
631 ntoskrnl_unicode_to_ascii(src->us_buf, dest->as_buf,
634 return (STATUS_SUCCESS);
638 RtlAnsiStringToUnicodeString(unicode_string *dest, ansi_string *src,
641 if (dest == NULL || src == NULL)
642 return (STATUS_INVALID_PARAMETER);
644 if (allocate == TRUE) {
645 dest->us_buf = ExAllocatePoolWithTag(NonPagedPool,
647 if (dest->us_buf == NULL)
648 return (STATUS_INSUFFICIENT_RESOURCES);
649 dest->us_len = dest->us_maxlen = strlen(src->as_buf) * 2;
651 dest->us_len = src->as_len * 2; /* XXX */
652 if (dest->us_maxlen < dest->us_len)
653 dest->us_len = dest->us_maxlen;
656 ntoskrnl_ascii_to_unicode(src->as_buf, dest->us_buf,
659 return (STATUS_SUCCESS);
663 ExAllocatePoolWithTag(pooltype, len, tag)
670 buf = malloc(len, M_DEVBUF, M_NOWAIT|M_ZERO);
678 ExFreePoolWithTag(buf, tag)
693 IoAllocateDriverObjectExtension(drv, clid, extlen, ext)
699 custom_extension *ce;
701 ce = ExAllocatePoolWithTag(NonPagedPool, sizeof(custom_extension)
705 return (STATUS_INSUFFICIENT_RESOURCES);
708 InsertTailList((&drv->dro_driverext->dre_usrext), (&ce->ce_list));
710 *ext = (void *)(ce + 1);
712 return (STATUS_SUCCESS);
716 IoGetDriverObjectExtension(drv, clid)
721 custom_extension *ce;
724 * Sanity check. Our dummy bus drivers don't have
725 * any driver extentions.
728 if (drv->dro_driverext == NULL)
731 e = drv->dro_driverext->dre_usrext.nle_flink;
732 while (e != &drv->dro_driverext->dre_usrext) {
733 ce = (custom_extension *)e;
734 if (ce->ce_clid == clid)
735 return ((void *)(ce + 1));
744 IoCreateDevice(driver_object *drv, uint32_t devextlen, unicode_string *devname,
745 uint32_t devtype, uint32_t devchars, uint8_t exclusive,
746 device_object **newdev)
750 dev = ExAllocatePoolWithTag(NonPagedPool, sizeof(device_object), 0);
752 return (STATUS_INSUFFICIENT_RESOURCES);
754 dev->do_type = devtype;
755 dev->do_drvobj = drv;
756 dev->do_currirp = NULL;
760 dev->do_devext = ExAllocatePoolWithTag(NonPagedPool,
763 if (dev->do_devext == NULL) {
765 return (STATUS_INSUFFICIENT_RESOURCES);
768 bzero(dev->do_devext, devextlen);
770 dev->do_devext = NULL;
772 dev->do_size = sizeof(device_object) + devextlen;
774 dev->do_attacheddev = NULL;
775 dev->do_nextdev = NULL;
776 dev->do_devtype = devtype;
777 dev->do_stacksize = 1;
778 dev->do_alignreq = 1;
779 dev->do_characteristics = devchars;
780 dev->do_iotimer = NULL;
781 KeInitializeEvent(&dev->do_devlock, EVENT_TYPE_SYNC, TRUE);
784 * Vpd is used for disk/tape devices,
785 * but we don't support those. (Yet.)
789 dev->do_devobj_ext = ExAllocatePoolWithTag(NonPagedPool,
790 sizeof(devobj_extension), 0);
792 if (dev->do_devobj_ext == NULL) {
793 if (dev->do_devext != NULL)
794 ExFreePool(dev->do_devext);
796 return (STATUS_INSUFFICIENT_RESOURCES);
799 dev->do_devobj_ext->dve_type = 0;
800 dev->do_devobj_ext->dve_size = sizeof(devobj_extension);
801 dev->do_devobj_ext->dve_devobj = dev;
804 * Attach this device to the driver object's list
805 * of devices. Note: this is not the same as attaching
806 * the device to the device stack. The driver's AddDevice
807 * routine must explicitly call IoAddDeviceToDeviceStack()
811 if (drv->dro_devobj == NULL) {
812 drv->dro_devobj = dev;
813 dev->do_nextdev = NULL;
815 dev->do_nextdev = drv->dro_devobj;
816 drv->dro_devobj = dev;
821 return (STATUS_SUCCESS);
833 if (dev->do_devobj_ext != NULL)
834 ExFreePool(dev->do_devobj_ext);
836 if (dev->do_devext != NULL)
837 ExFreePool(dev->do_devext);
839 /* Unlink the device from the driver's device list. */
841 prev = dev->do_drvobj->dro_devobj;
843 dev->do_drvobj->dro_devobj = dev->do_nextdev;
845 while (prev->do_nextdev != dev)
846 prev = prev->do_nextdev;
847 prev->do_nextdev = dev->do_nextdev;
854 IoGetAttachedDevice(dev)
864 while (d->do_attacheddev != NULL)
865 d = d->do_attacheddev;
871 IoBuildSynchronousFsdRequest(func, dobj, buf, len, off, event, status)
878 io_status_block *status;
882 ip = IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status);
885 ip->irp_usrevent = event;
891 IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status)
897 io_status_block *status;
900 io_stack_location *sl;
902 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
906 ip->irp_usriostat = status;
907 ip->irp_tail.irp_overlay.irp_thread = NULL;
909 sl = IoGetNextIrpStackLocation(ip);
910 sl->isl_major = func;
914 sl->isl_devobj = dobj;
915 sl->isl_fileobj = NULL;
916 sl->isl_completionfunc = NULL;
918 ip->irp_userbuf = buf;
920 if (dobj->do_flags & DO_BUFFERED_IO) {
921 ip->irp_assoc.irp_sysbuf =
922 ExAllocatePoolWithTag(NonPagedPool, len, 0);
923 if (ip->irp_assoc.irp_sysbuf == NULL) {
927 bcopy(buf, ip->irp_assoc.irp_sysbuf, len);
930 if (dobj->do_flags & DO_DIRECT_IO) {
931 ip->irp_mdl = IoAllocateMdl(buf, len, FALSE, FALSE, ip);
932 if (ip->irp_mdl == NULL) {
933 if (ip->irp_assoc.irp_sysbuf != NULL)
934 ExFreePool(ip->irp_assoc.irp_sysbuf);
938 ip->irp_userbuf = NULL;
939 ip->irp_assoc.irp_sysbuf = NULL;
942 if (func == IRP_MJ_READ) {
943 sl->isl_parameters.isl_read.isl_len = len;
945 sl->isl_parameters.isl_read.isl_byteoff = *off;
947 sl->isl_parameters.isl_read.isl_byteoff = 0;
950 if (func == IRP_MJ_WRITE) {
951 sl->isl_parameters.isl_write.isl_len = len;
953 sl->isl_parameters.isl_write.isl_byteoff = *off;
955 sl->isl_parameters.isl_write.isl_byteoff = 0;
962 IoBuildDeviceIoControlRequest(uint32_t iocode, device_object *dobj, void *ibuf,
963 uint32_t ilen, void *obuf, uint32_t olen, uint8_t isinternal,
964 nt_kevent *event, io_status_block *status)
967 io_stack_location *sl;
970 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
973 ip->irp_usrevent = event;
974 ip->irp_usriostat = status;
975 ip->irp_tail.irp_overlay.irp_thread = NULL;
977 sl = IoGetNextIrpStackLocation(ip);
978 sl->isl_major = isinternal == TRUE ?
979 IRP_MJ_INTERNAL_DEVICE_CONTROL : IRP_MJ_DEVICE_CONTROL;
983 sl->isl_devobj = dobj;
984 sl->isl_fileobj = NULL;
985 sl->isl_completionfunc = NULL;
986 sl->isl_parameters.isl_ioctl.isl_iocode = iocode;
987 sl->isl_parameters.isl_ioctl.isl_ibuflen = ilen;
988 sl->isl_parameters.isl_ioctl.isl_obuflen = olen;
990 switch(IO_METHOD(iocode)) {
991 case METHOD_BUFFERED:
997 ip->irp_assoc.irp_sysbuf =
998 ExAllocatePoolWithTag(NonPagedPool, buflen, 0);
999 if (ip->irp_assoc.irp_sysbuf == NULL) {
1004 if (ilen && ibuf != NULL) {
1005 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1006 bzero((char *)ip->irp_assoc.irp_sysbuf + ilen,
1009 bzero(ip->irp_assoc.irp_sysbuf, ilen);
1010 ip->irp_userbuf = obuf;
1012 case METHOD_IN_DIRECT:
1013 case METHOD_OUT_DIRECT:
1014 if (ilen && ibuf != NULL) {
1015 ip->irp_assoc.irp_sysbuf =
1016 ExAllocatePoolWithTag(NonPagedPool, ilen, 0);
1017 if (ip->irp_assoc.irp_sysbuf == NULL) {
1021 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1023 if (olen && obuf != NULL) {
1024 ip->irp_mdl = IoAllocateMdl(obuf, olen,
1027 * Normally we would MmProbeAndLockPages()
1028 * here, but we don't have to in our
1033 case METHOD_NEITHER:
1034 ip->irp_userbuf = obuf;
1035 sl->isl_parameters.isl_ioctl.isl_type3ibuf = ibuf;
1042 * Ideally, we should associate this IRP with the calling
1050 IoAllocateIrp(uint8_t stsize, uint8_t chargequota)
1054 i = ExAllocatePoolWithTag(NonPagedPool, IoSizeOfIrp(stsize), 0);
1058 IoInitializeIrp(i, IoSizeOfIrp(stsize), stsize);
1064 IoMakeAssociatedIrp(irp *ip, uint8_t stsize)
1068 associrp = IoAllocateIrp(stsize, FALSE);
1069 if (associrp == NULL)
1072 mtx_lock(&ntoskrnl_dispatchlock);
1073 associrp->irp_flags |= IRP_ASSOCIATED_IRP;
1074 associrp->irp_tail.irp_overlay.irp_thread =
1075 ip->irp_tail.irp_overlay.irp_thread;
1076 associrp->irp_assoc.irp_master = ip;
1077 mtx_unlock(&ntoskrnl_dispatchlock);
1090 IoInitializeIrp(irp *io, uint16_t psize, uint8_t ssize)
1092 bzero((char *)io, IoSizeOfIrp(ssize));
1093 io->irp_size = psize;
1094 io->irp_stackcnt = ssize;
1095 io->irp_currentstackloc = ssize;
1096 InitializeListHead(&io->irp_thlist);
1097 io->irp_tail.irp_overlay.irp_csl =
1098 (io_stack_location *)(io + 1) + ssize;
1102 IoReuseIrp(ip, status)
1108 allocflags = ip->irp_allocflags;
1109 IoInitializeIrp(ip, ip->irp_size, ip->irp_stackcnt);
1110 ip->irp_iostat.isb_status = status;
1111 ip->irp_allocflags = allocflags;
1115 IoAcquireCancelSpinLock(uint8_t *irql)
1117 KeAcquireSpinLock(&ntoskrnl_cancellock, irql);
1121 IoReleaseCancelSpinLock(uint8_t irql)
1123 KeReleaseSpinLock(&ntoskrnl_cancellock, irql);
1127 IoCancelIrp(irp *ip)
1132 IoAcquireCancelSpinLock(&cancelirql);
1133 cfunc = IoSetCancelRoutine(ip, NULL);
1134 ip->irp_cancel = TRUE;
1135 if (cfunc == NULL) {
1136 IoReleaseCancelSpinLock(cancelirql);
1139 ip->irp_cancelirql = cancelirql;
1140 MSCALL2(cfunc, IoGetCurrentIrpStackLocation(ip)->isl_devobj, ip);
1141 return (uint8_t)IoSetCancelValue(ip, TRUE);
1145 IofCallDriver(dobj, ip)
1146 device_object *dobj;
1149 driver_object *drvobj;
1150 io_stack_location *sl;
1152 driver_dispatch disp;
1154 drvobj = dobj->do_drvobj;
1156 if (ip->irp_currentstackloc <= 0)
1157 panic("IoCallDriver(): out of stack locations");
1159 IoSetNextIrpStackLocation(ip);
1160 sl = IoGetCurrentIrpStackLocation(ip);
1162 sl->isl_devobj = dobj;
1164 disp = drvobj->dro_dispatch[sl->isl_major];
1165 status = MSCALL2(disp, dobj, ip);
1171 IofCompleteRequest(irp *ip, uint8_t prioboost)
1174 device_object *dobj;
1175 io_stack_location *sl;
1178 KASSERT(ip->irp_iostat.isb_status != STATUS_PENDING,
1179 ("incorrect IRP(%p) status (STATUS_PENDING)", ip));
1181 sl = IoGetCurrentIrpStackLocation(ip);
1182 IoSkipCurrentIrpStackLocation(ip);
1185 if (sl->isl_ctl & SL_PENDING_RETURNED)
1186 ip->irp_pendingreturned = TRUE;
1188 if (ip->irp_currentstackloc != (ip->irp_stackcnt + 1))
1189 dobj = IoGetCurrentIrpStackLocation(ip)->isl_devobj;
1193 if (sl->isl_completionfunc != NULL &&
1194 ((ip->irp_iostat.isb_status == STATUS_SUCCESS &&
1195 sl->isl_ctl & SL_INVOKE_ON_SUCCESS) ||
1196 (ip->irp_iostat.isb_status != STATUS_SUCCESS &&
1197 sl->isl_ctl & SL_INVOKE_ON_ERROR) ||
1198 (ip->irp_cancel == TRUE &&
1199 sl->isl_ctl & SL_INVOKE_ON_CANCEL))) {
1200 cf = sl->isl_completionfunc;
1201 status = MSCALL3(cf, dobj, ip, sl->isl_completionctx);
1202 if (status == STATUS_MORE_PROCESSING_REQUIRED)
1205 if ((ip->irp_currentstackloc <= ip->irp_stackcnt) &&
1206 (ip->irp_pendingreturned == TRUE))
1207 IoMarkIrpPending(ip);
1210 /* move to the next. */
1211 IoSkipCurrentIrpStackLocation(ip);
1213 } while (ip->irp_currentstackloc <= (ip->irp_stackcnt + 1));
1215 if (ip->irp_usriostat != NULL)
1216 *ip->irp_usriostat = ip->irp_iostat;
1217 if (ip->irp_usrevent != NULL)
1218 KeSetEvent(ip->irp_usrevent, prioboost, FALSE);
1220 /* Handle any associated IRPs. */
1222 if (ip->irp_flags & IRP_ASSOCIATED_IRP) {
1223 uint32_t masterirpcnt;
1227 masterirp = ip->irp_assoc.irp_master;
1229 InterlockedDecrement(&masterirp->irp_assoc.irp_irpcnt);
1231 while ((m = ip->irp_mdl) != NULL) {
1232 ip->irp_mdl = m->mdl_next;
1236 if (masterirpcnt == 0)
1237 IoCompleteRequest(masterirp, IO_NO_INCREMENT);
1241 /* With any luck, these conditions will never arise. */
1243 if (ip->irp_flags & IRP_PAGING_IO) {
1244 if (ip->irp_mdl != NULL)
1245 IoFreeMdl(ip->irp_mdl);
1259 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1260 l = ntoskrnl_intlist.nle_flink;
1261 while (l != &ntoskrnl_intlist) {
1262 iobj = CONTAINING_RECORD(l, kinterrupt, ki_list);
1263 claimed = MSCALL2(iobj->ki_svcfunc, iobj, iobj->ki_svcctx);
1264 if (claimed == TRUE)
1268 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1272 KeAcquireInterruptSpinLock(iobj)
1276 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1281 KeReleaseInterruptSpinLock(kinterrupt *iobj, uint8_t irql)
1283 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1287 KeSynchronizeExecution(iobj, syncfunc, syncctx)
1294 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1295 MSCALL1(syncfunc, syncctx);
1296 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1302 * IoConnectInterrupt() is passed only the interrupt vector and
1303 * irql that a device wants to use, but no device-specific tag
1304 * of any kind. This conflicts rather badly with FreeBSD's
1305 * bus_setup_intr(), which needs the device_t for the device
1306 * requesting interrupt delivery. In order to bypass this
1307 * inconsistency, we implement a second level of interrupt
1308 * dispatching on top of bus_setup_intr(). All devices use
1309 * ntoskrnl_intr() as their ISR, and any device requesting
1310 * interrupts will be registered with ntoskrnl_intr()'s interrupt
1311 * dispatch list. When an interrupt arrives, we walk the list
1312 * and invoke all the registered ISRs. This effectively makes all
1313 * interrupts shared, but it's the only way to duplicate the
1314 * semantics of IoConnectInterrupt() and IoDisconnectInterrupt() properly.
1318 IoConnectInterrupt(kinterrupt **iobj, void *svcfunc, void *svcctx,
1319 kspin_lock *lock, uint32_t vector, uint8_t irql, uint8_t syncirql,
1320 uint8_t imode, uint8_t shared, uint32_t affinity, uint8_t savefloat)
1324 *iobj = ExAllocatePoolWithTag(NonPagedPool, sizeof(kinterrupt), 0);
1326 return (STATUS_INSUFFICIENT_RESOURCES);
1328 (*iobj)->ki_svcfunc = svcfunc;
1329 (*iobj)->ki_svcctx = svcctx;
1332 KeInitializeSpinLock(&(*iobj)->ki_lock_priv);
1333 (*iobj)->ki_lock = &(*iobj)->ki_lock_priv;
1335 (*iobj)->ki_lock = lock;
1337 KeAcquireSpinLock(&ntoskrnl_intlock, &curirql);
1338 InsertHeadList((&ntoskrnl_intlist), (&(*iobj)->ki_list));
1339 KeReleaseSpinLock(&ntoskrnl_intlock, curirql);
1341 return (STATUS_SUCCESS);
1345 IoDisconnectInterrupt(iobj)
1353 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1354 RemoveEntryList((&iobj->ki_list));
1355 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1361 IoAttachDeviceToDeviceStack(src, dst)
1365 device_object *attached;
1367 mtx_lock(&ntoskrnl_dispatchlock);
1368 attached = IoGetAttachedDevice(dst);
1369 attached->do_attacheddev = src;
1370 src->do_attacheddev = NULL;
1371 src->do_stacksize = attached->do_stacksize + 1;
1372 mtx_unlock(&ntoskrnl_dispatchlock);
1378 IoDetachDevice(topdev)
1379 device_object *topdev;
1381 device_object *tail;
1383 mtx_lock(&ntoskrnl_dispatchlock);
1385 /* First, break the chain. */
1386 tail = topdev->do_attacheddev;
1388 mtx_unlock(&ntoskrnl_dispatchlock);
1391 topdev->do_attacheddev = tail->do_attacheddev;
1392 topdev->do_refcnt--;
1394 /* Now reduce the stacksize count for the takm_il objects. */
1396 tail = topdev->do_attacheddev;
1397 while (tail != NULL) {
1398 tail->do_stacksize--;
1399 tail = tail->do_attacheddev;
1402 mtx_unlock(&ntoskrnl_dispatchlock);
1406 * For the most part, an object is considered signalled if
1407 * dh_sigstate == TRUE. The exception is for mutant objects
1408 * (mutexes), where the logic works like this:
1410 * - If the thread already owns the object and sigstate is
1411 * less than or equal to 0, then the object is considered
1412 * signalled (recursive acquisition).
1413 * - If dh_sigstate == 1, the object is also considered
1418 ntoskrnl_is_signalled(obj, td)
1419 nt_dispatch_header *obj;
1424 if (obj->dh_type == DISP_TYPE_MUTANT) {
1425 km = (kmutant *)obj;
1426 if ((obj->dh_sigstate <= 0 && km->km_ownerthread == td) ||
1427 obj->dh_sigstate == 1)
1432 if (obj->dh_sigstate > 0)
1438 ntoskrnl_satisfy_wait(obj, td)
1439 nt_dispatch_header *obj;
1444 switch (obj->dh_type) {
1445 case DISP_TYPE_MUTANT:
1446 km = (struct kmutant *)obj;
1449 * If sigstate reaches 0, the mutex is now
1450 * non-signalled (the new thread owns it).
1452 if (obj->dh_sigstate == 0) {
1453 km->km_ownerthread = td;
1454 if (km->km_abandoned == TRUE)
1455 km->km_abandoned = FALSE;
1458 /* Synchronization objects get reset to unsignalled. */
1459 case DISP_TYPE_SYNCHRONIZATION_EVENT:
1460 case DISP_TYPE_SYNCHRONIZATION_TIMER:
1461 obj->dh_sigstate = 0;
1463 case DISP_TYPE_SEMAPHORE:
1472 ntoskrnl_satisfy_multiple_waits(wb)
1479 td = wb->wb_kthread;
1482 ntoskrnl_satisfy_wait(wb->wb_object, td);
1483 cur->wb_awakened = TRUE;
1485 } while (cur != wb);
1488 /* Always called with dispatcher lock held. */
1490 ntoskrnl_waittest(obj, increment)
1491 nt_dispatch_header *obj;
1494 wait_block *w, *next;
1501 * Once an object has been signalled, we walk its list of
1502 * wait blocks. If a wait block can be awakened, then satisfy
1503 * waits as necessary and wake the thread.
1505 * The rules work like this:
1507 * If a wait block is marked as WAITTYPE_ANY, then
1508 * we can satisfy the wait conditions on the current
1509 * object and wake the thread right away. Satisfying
1510 * the wait also has the effect of breaking us out
1511 * of the search loop.
1513 * If the object is marked as WAITTYLE_ALL, then the
1514 * wait block will be part of a circularly linked
1515 * list of wait blocks belonging to a waiting thread
1516 * that's sleeping in KeWaitForMultipleObjects(). In
1517 * order to wake the thread, all the objects in the
1518 * wait list must be in the signalled state. If they
1519 * are, we then satisfy all of them and wake the
1524 e = obj->dh_waitlisthead.nle_flink;
1526 while (e != &obj->dh_waitlisthead && obj->dh_sigstate > 0) {
1527 w = CONTAINING_RECORD(e, wait_block, wb_waitlist);
1531 if (w->wb_waittype == WAITTYPE_ANY) {
1533 * Thread can be awakened if
1534 * any wait is satisfied.
1536 ntoskrnl_satisfy_wait(obj, td);
1538 w->wb_awakened = TRUE;
1541 * Thread can only be woken up
1542 * if all waits are satisfied.
1543 * If the thread is waiting on multiple
1544 * objects, they should all be linked
1545 * through the wb_next pointers in the
1551 if (ntoskrnl_is_signalled(obj, td) == FALSE) {
1555 next = next->wb_next;
1557 ntoskrnl_satisfy_multiple_waits(w);
1560 if (satisfied == TRUE)
1561 cv_broadcastpri(&we->we_cv,
1562 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
1563 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
1570 * Return the number of 100 nanosecond intervals since
1571 * January 1, 1601. (?!?!)
1580 *tval = (uint64_t)ts.tv_nsec / 100 + (uint64_t)ts.tv_sec * 10000000 +
1581 11644473600 * 10000000; /* 100ns ticks from 1601 to 1970 */
1585 KeQuerySystemTime(current_time)
1586 uint64_t *current_time;
1588 ntoskrnl_time(current_time);
1595 getmicrouptime(&tv);
1601 * KeWaitForSingleObject() is a tricky beast, because it can be used
1602 * with several different object types: semaphores, timers, events,
1603 * mutexes and threads. Semaphores don't appear very often, but the
1604 * other object types are quite common. KeWaitForSingleObject() is
1605 * what's normally used to acquire a mutex, and it can be used to
1606 * wait for a thread termination.
1608 * The Windows NDIS API is implemented in terms of Windows kernel
1609 * primitives, and some of the object manipulation is duplicated in
1610 * NDIS. For example, NDIS has timers and events, which are actually
1611 * Windows kevents and ktimers. Now, you're supposed to only use the
1612 * NDIS variants of these objects within the confines of the NDIS API,
1613 * but there are some naughty developers out there who will use
1614 * KeWaitForSingleObject() on NDIS timer and event objects, so we
1615 * have to support that as well. Conseqently, our NDIS timer and event
1616 * code has to be closely tied into our ntoskrnl timer and event code,
1617 * just as it is in Windows.
1619 * KeWaitForSingleObject() may do different things for different kinds
1622 * - For events, we check if the event has been signalled. If the
1623 * event is already in the signalled state, we just return immediately,
1624 * otherwise we wait for it to be set to the signalled state by someone
1625 * else calling KeSetEvent(). Events can be either synchronization or
1626 * notification events.
1628 * - For timers, if the timer has already fired and the timer is in
1629 * the signalled state, we just return, otherwise we wait on the
1630 * timer. Unlike an event, timers get signalled automatically when
1631 * they expire rather than someone having to trip them manually.
1632 * Timers initialized with KeInitializeTimer() are always notification
1633 * events: KeInitializeTimerEx() lets you initialize a timer as
1634 * either a notification or synchronization event.
1636 * - For mutexes, we try to acquire the mutex and if we can't, we wait
1637 * on the mutex until it's available and then grab it. When a mutex is
1638 * released, it enters the signalled state, which wakes up one of the
1639 * threads waiting to acquire it. Mutexes are always synchronization
1642 * - For threads, the only thing we do is wait until the thread object
1643 * enters a signalled state, which occurs when the thread terminates.
1644 * Threads are always notification events.
1646 * A notification event wakes up all threads waiting on an object. A
1647 * synchronization event wakes up just one. Also, a synchronization event
1648 * is auto-clearing, which means we automatically set the event back to
1649 * the non-signalled state once the wakeup is done.
1653 KeWaitForSingleObject(void *arg, uint32_t reason, uint32_t mode,
1654 uint8_t alertable, int64_t *duetime)
1657 struct thread *td = curthread;
1662 nt_dispatch_header *obj;
1667 return (STATUS_INVALID_PARAMETER);
1669 mtx_lock(&ntoskrnl_dispatchlock);
1671 cv_init(&we.we_cv, "KeWFS");
1675 * Check to see if this object is already signalled,
1676 * and just return without waiting if it is.
1678 if (ntoskrnl_is_signalled(obj, td) == TRUE) {
1679 /* Sanity check the signal state value. */
1680 if (obj->dh_sigstate != INT32_MIN) {
1681 ntoskrnl_satisfy_wait(obj, curthread);
1682 mtx_unlock(&ntoskrnl_dispatchlock);
1683 return (STATUS_SUCCESS);
1686 * There's a limit to how many times we can
1687 * recursively acquire a mutant. If we hit
1688 * the limit, something is very wrong.
1690 if (obj->dh_type == DISP_TYPE_MUTANT) {
1691 mtx_unlock(&ntoskrnl_dispatchlock);
1692 panic("mutant limit exceeded");
1697 bzero((char *)&w, sizeof(wait_block));
1700 w.wb_waittype = WAITTYPE_ANY;
1703 w.wb_awakened = FALSE;
1704 w.wb_oldpri = td->td_priority;
1706 InsertTailList((&obj->dh_waitlisthead), (&w.wb_waitlist));
1709 * The timeout value is specified in 100 nanosecond units
1710 * and can be a positive or negative number. If it's positive,
1711 * then the duetime is absolute, and we need to convert it
1712 * to an absolute offset relative to now in order to use it.
1713 * If it's negative, then the duetime is relative and we
1714 * just have to convert the units.
1717 if (duetime != NULL) {
1719 tv.tv_sec = - (*duetime) / 10000000;
1720 tv.tv_usec = (- (*duetime) / 10) -
1721 (tv.tv_sec * 1000000);
1723 ntoskrnl_time(&curtime);
1724 if (*duetime < curtime)
1725 tv.tv_sec = tv.tv_usec = 0;
1727 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1728 tv.tv_usec = ((*duetime) - curtime) / 10 -
1729 (tv.tv_sec * 1000000);
1734 if (duetime == NULL)
1735 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1737 error = cv_timedwait(&we.we_cv,
1738 &ntoskrnl_dispatchlock, tvtohz(&tv));
1740 RemoveEntryList(&w.wb_waitlist);
1742 cv_destroy(&we.we_cv);
1744 /* We timed out. Leave the object alone and return status. */
1746 if (error == EWOULDBLOCK) {
1747 mtx_unlock(&ntoskrnl_dispatchlock);
1748 return (STATUS_TIMEOUT);
1751 mtx_unlock(&ntoskrnl_dispatchlock);
1753 return (STATUS_SUCCESS);
1755 return (KeWaitForMultipleObjects(1, &obj, WAITTYPE_ALL, reason,
1756 mode, alertable, duetime, &w));
1761 KeWaitForMultipleObjects(uint32_t cnt, nt_dispatch_header *obj[], uint32_t wtype,
1762 uint32_t reason, uint32_t mode, uint8_t alertable, int64_t *duetime,
1763 wait_block *wb_array)
1765 struct thread *td = curthread;
1766 wait_block *whead, *w;
1767 wait_block _wb_array[MAX_WAIT_OBJECTS];
1768 nt_dispatch_header *cur;
1770 int i, wcnt = 0, error = 0;
1772 struct timespec t1, t2;
1773 uint32_t status = STATUS_SUCCESS;
1776 if (cnt > MAX_WAIT_OBJECTS)
1777 return (STATUS_INVALID_PARAMETER);
1778 if (cnt > THREAD_WAIT_OBJECTS && wb_array == NULL)
1779 return (STATUS_INVALID_PARAMETER);
1781 mtx_lock(&ntoskrnl_dispatchlock);
1783 cv_init(&we.we_cv, "KeWFM");
1786 if (wb_array == NULL)
1791 bzero((char *)whead, sizeof(wait_block) * cnt);
1793 /* First pass: see if we can satisfy any waits immediately. */
1798 for (i = 0; i < cnt; i++) {
1799 InsertTailList((&obj[i]->dh_waitlisthead),
1802 w->wb_object = obj[i];
1803 w->wb_waittype = wtype;
1805 w->wb_awakened = FALSE;
1806 w->wb_oldpri = td->td_priority;
1810 if (ntoskrnl_is_signalled(obj[i], td)) {
1812 * There's a limit to how many times
1813 * we can recursively acquire a mutant.
1814 * If we hit the limit, something
1817 if (obj[i]->dh_sigstate == INT32_MIN &&
1818 obj[i]->dh_type == DISP_TYPE_MUTANT) {
1819 mtx_unlock(&ntoskrnl_dispatchlock);
1820 panic("mutant limit exceeded");
1824 * If this is a WAITTYPE_ANY wait, then
1825 * satisfy the waited object and exit
1829 if (wtype == WAITTYPE_ANY) {
1830 ntoskrnl_satisfy_wait(obj[i], td);
1831 status = STATUS_WAIT_0 + i;
1836 w->wb_object = NULL;
1837 RemoveEntryList(&w->wb_waitlist);
1843 * If this is a WAITTYPE_ALL wait and all objects are
1844 * already signalled, satisfy the waits and exit now.
1847 if (wtype == WAITTYPE_ALL && wcnt == 0) {
1848 for (i = 0; i < cnt; i++)
1849 ntoskrnl_satisfy_wait(obj[i], td);
1850 status = STATUS_SUCCESS;
1855 * Create a circular waitblock list. The waitcount
1856 * must always be non-zero when we get here.
1859 (w - 1)->wb_next = whead;
1861 /* Wait on any objects that aren't yet signalled. */
1863 /* Calculate timeout, if any. */
1865 if (duetime != NULL) {
1867 tv.tv_sec = - (*duetime) / 10000000;
1868 tv.tv_usec = (- (*duetime) / 10) -
1869 (tv.tv_sec * 1000000);
1871 ntoskrnl_time(&curtime);
1872 if (*duetime < curtime)
1873 tv.tv_sec = tv.tv_usec = 0;
1875 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1876 tv.tv_usec = ((*duetime) - curtime) / 10 -
1877 (tv.tv_sec * 1000000);
1885 if (duetime == NULL)
1886 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1888 error = cv_timedwait(&we.we_cv,
1889 &ntoskrnl_dispatchlock, tvtohz(&tv));
1891 /* Wait with timeout expired. */
1894 status = STATUS_TIMEOUT;
1900 /* See what's been signalled. */
1905 if (ntoskrnl_is_signalled(cur, td) == TRUE ||
1906 w->wb_awakened == TRUE) {
1907 /* Sanity check the signal state value. */
1908 if (cur->dh_sigstate == INT32_MIN &&
1909 cur->dh_type == DISP_TYPE_MUTANT) {
1910 mtx_unlock(&ntoskrnl_dispatchlock);
1911 panic("mutant limit exceeded");
1914 if (wtype == WAITTYPE_ANY) {
1915 status = w->wb_waitkey &
1921 } while (w != whead);
1924 * If all objects have been signalled, or if this
1925 * is a WAITTYPE_ANY wait and we were woke up by
1926 * someone, we can bail.
1930 status = STATUS_SUCCESS;
1935 * If this is WAITTYPE_ALL wait, and there's still
1936 * objects that haven't been signalled, deduct the
1937 * time that's elapsed so far from the timeout and
1938 * wait again (or continue waiting indefinitely if
1939 * there's no timeout).
1942 if (duetime != NULL) {
1943 tv.tv_sec -= (t2.tv_sec - t1.tv_sec);
1944 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);
1958 mtx_unlock(&ntoskrnl_dispatchlock);
1964 WRITE_REGISTER_USHORT(uint16_t *reg, uint16_t val)
1966 bus_space_write_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1970 READ_REGISTER_USHORT(reg)
1973 return (bus_space_read_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1977 WRITE_REGISTER_ULONG(reg, val)
1981 bus_space_write_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1985 READ_REGISTER_ULONG(reg)
1988 return (bus_space_read_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1992 READ_REGISTER_UCHAR(uint8_t *reg)
1994 return (bus_space_read_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1998 WRITE_REGISTER_UCHAR(uint8_t *reg, uint8_t val)
2000 bus_space_write_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
2052 _allshl(int64_t a, uint8_t b)
2058 _aullshl(uint64_t a, uint8_t b)
2064 _allshr(int64_t a, uint8_t b)
2070 _aullshr(uint64_t a, uint8_t b)
2075 static slist_entry *
2076 ntoskrnl_pushsl(head, entry)
2080 slist_entry *oldhead;
2082 oldhead = head->slh_list.slh_next;
2083 entry->sl_next = head->slh_list.slh_next;
2084 head->slh_list.slh_next = entry;
2085 head->slh_list.slh_depth++;
2086 head->slh_list.slh_seq++;
2092 InitializeSListHead(head)
2095 memset(head, 0, sizeof(*head));
2098 static slist_entry *
2099 ntoskrnl_popsl(head)
2104 first = head->slh_list.slh_next;
2105 if (first != NULL) {
2106 head->slh_list.slh_next = first->sl_next;
2107 head->slh_list.slh_depth--;
2108 head->slh_list.slh_seq++;
2115 * We need this to make lookaside lists work for amd64.
2116 * We pass a pointer to ExAllocatePoolWithTag() the lookaside
2117 * list structure. For amd64 to work right, this has to be a
2118 * pointer to the wrapped version of the routine, not the
2119 * original. Letting the Windows driver invoke the original
2120 * function directly will result in a convention calling
2121 * mismatch and a pretty crash. On x86, this effectively
2122 * becomes a no-op since ipt_func and ipt_wrap are the same.
2126 ntoskrnl_findwrap(func)
2129 image_patch_table *patch;
2131 patch = ntoskrnl_functbl;
2132 while (patch->ipt_func != NULL) {
2133 if ((funcptr)patch->ipt_func == func)
2134 return ((funcptr)patch->ipt_wrap);
2142 ExInitializePagedLookasideList(paged_lookaside_list *lookaside,
2143 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2144 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2146 bzero((char *)lookaside, sizeof(paged_lookaside_list));
2148 if (size < sizeof(slist_entry))
2149 lookaside->nll_l.gl_size = sizeof(slist_entry);
2151 lookaside->nll_l.gl_size = size;
2152 lookaside->nll_l.gl_tag = tag;
2153 if (allocfunc == NULL)
2154 lookaside->nll_l.gl_allocfunc =
2155 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2157 lookaside->nll_l.gl_allocfunc = allocfunc;
2159 if (freefunc == NULL)
2160 lookaside->nll_l.gl_freefunc =
2161 ntoskrnl_findwrap((funcptr)ExFreePool);
2163 lookaside->nll_l.gl_freefunc = freefunc;
2166 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2169 lookaside->nll_l.gl_type = NonPagedPool;
2170 lookaside->nll_l.gl_depth = depth;
2171 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2175 ExDeletePagedLookasideList(lookaside)
2176 paged_lookaside_list *lookaside;
2179 void (*freefunc)(void *);
2181 freefunc = lookaside->nll_l.gl_freefunc;
2182 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2183 MSCALL1(freefunc, buf);
2187 ExInitializeNPagedLookasideList(npaged_lookaside_list *lookaside,
2188 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2189 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2191 bzero((char *)lookaside, sizeof(npaged_lookaside_list));
2193 if (size < sizeof(slist_entry))
2194 lookaside->nll_l.gl_size = sizeof(slist_entry);
2196 lookaside->nll_l.gl_size = size;
2197 lookaside->nll_l.gl_tag = tag;
2198 if (allocfunc == NULL)
2199 lookaside->nll_l.gl_allocfunc =
2200 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2202 lookaside->nll_l.gl_allocfunc = allocfunc;
2204 if (freefunc == NULL)
2205 lookaside->nll_l.gl_freefunc =
2206 ntoskrnl_findwrap((funcptr)ExFreePool);
2208 lookaside->nll_l.gl_freefunc = freefunc;
2211 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2214 lookaside->nll_l.gl_type = NonPagedPool;
2215 lookaside->nll_l.gl_depth = depth;
2216 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2220 ExDeleteNPagedLookasideList(lookaside)
2221 npaged_lookaside_list *lookaside;
2224 void (*freefunc)(void *);
2226 freefunc = lookaside->nll_l.gl_freefunc;
2227 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2228 MSCALL1(freefunc, buf);
2232 InterlockedPushEntrySList(head, entry)
2236 slist_entry *oldhead;
2238 mtx_lock_spin(&ntoskrnl_interlock);
2239 oldhead = ntoskrnl_pushsl(head, entry);
2240 mtx_unlock_spin(&ntoskrnl_interlock);
2246 InterlockedPopEntrySList(head)
2251 mtx_lock_spin(&ntoskrnl_interlock);
2252 first = ntoskrnl_popsl(head);
2253 mtx_unlock_spin(&ntoskrnl_interlock);
2258 static slist_entry *
2259 ExInterlockedPushEntrySList(head, entry, lock)
2264 return (InterlockedPushEntrySList(head, entry));
2267 static slist_entry *
2268 ExInterlockedPopEntrySList(head, lock)
2272 return (InterlockedPopEntrySList(head));
2276 ExQueryDepthSList(head)
2281 mtx_lock_spin(&ntoskrnl_interlock);
2282 depth = head->slh_list.slh_depth;
2283 mtx_unlock_spin(&ntoskrnl_interlock);
2289 KeInitializeSpinLock(lock)
2297 KefAcquireSpinLockAtDpcLevel(lock)
2300 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2304 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0) {
2306 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2315 KefReleaseSpinLockFromDpcLevel(lock)
2318 atomic_store_rel_int((volatile u_int *)lock, 0);
2322 KeAcquireSpinLockRaiseToDpc(kspin_lock *lock)
2326 if (KeGetCurrentIrql() > DISPATCH_LEVEL)
2327 panic("IRQL_NOT_LESS_THAN_OR_EQUAL");
2329 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
2330 KeAcquireSpinLockAtDpcLevel(lock);
2336 KeAcquireSpinLockAtDpcLevel(kspin_lock *lock)
2338 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0)
2343 KeReleaseSpinLockFromDpcLevel(kspin_lock *lock)
2345 atomic_store_rel_int((volatile u_int *)lock, 0);
2347 #endif /* __i386__ */
2350 InterlockedExchange(dst, val)
2351 volatile uint32_t *dst;
2356 mtx_lock_spin(&ntoskrnl_interlock);
2359 mtx_unlock_spin(&ntoskrnl_interlock);
2365 InterlockedIncrement(addend)
2366 volatile uint32_t *addend;
2368 atomic_add_long((volatile u_long *)addend, 1);
2373 InterlockedDecrement(addend)
2374 volatile uint32_t *addend;
2376 atomic_subtract_long((volatile u_long *)addend, 1);
2381 ExInterlockedAddLargeStatistic(addend, inc)
2385 mtx_lock_spin(&ntoskrnl_interlock);
2387 mtx_unlock_spin(&ntoskrnl_interlock);
2391 IoAllocateMdl(void *vaddr, uint32_t len, uint8_t secondarybuf,
2392 uint8_t chargequota, irp *iopkt)
2397 if (MmSizeOfMdl(vaddr, len) > MDL_ZONE_SIZE)
2398 m = ExAllocatePoolWithTag(NonPagedPool,
2399 MmSizeOfMdl(vaddr, len), 0);
2401 m = uma_zalloc(mdl_zone, M_NOWAIT | M_ZERO);
2408 MmInitializeMdl(m, vaddr, len);
2411 * MmInitializMdl() clears the flags field, so we
2412 * have to set this here. If the MDL came from the
2413 * MDL UMA zone, tag it so we can release it to
2414 * the right place later.
2417 m->mdl_flags = MDL_ZONE_ALLOCED;
2419 if (iopkt != NULL) {
2420 if (secondarybuf == TRUE) {
2422 last = iopkt->irp_mdl;
2423 while (last->mdl_next != NULL)
2424 last = last->mdl_next;
2427 if (iopkt->irp_mdl != NULL)
2428 panic("leaking an MDL in IoAllocateMdl()");
2443 if (m->mdl_flags & MDL_ZONE_ALLOCED)
2444 uma_zfree(mdl_zone, m);
2450 MmAllocateContiguousMemory(size, highest)
2455 size_t pagelength = roundup(size, PAGE_SIZE);
2457 addr = ExAllocatePoolWithTag(NonPagedPool, pagelength, 0);
2463 MmAllocateContiguousMemorySpecifyCache(size, lowest, highest,
2464 boundary, cachetype)
2469 enum nt_caching_type cachetype;
2471 vm_memattr_t memattr;
2474 switch (cachetype) {
2476 memattr = VM_MEMATTR_UNCACHEABLE;
2478 case MmWriteCombined:
2479 memattr = VM_MEMATTR_WRITE_COMBINING;
2481 case MmNonCachedUnordered:
2482 memattr = VM_MEMATTR_UNCACHEABLE;
2485 case MmHardwareCoherentCached:
2488 memattr = VM_MEMATTR_DEFAULT;
2492 ret = (void *)kmem_alloc_contig(kernel_arena, size, M_ZERO | M_NOWAIT,
2493 lowest, highest, PAGE_SIZE, boundary, memattr);
2495 malloc_type_allocated(M_DEVBUF, round_page(size));
2500 MmFreeContiguousMemory(base)
2507 MmFreeContiguousMemorySpecifyCache(base, size, cachetype)
2510 enum nt_caching_type cachetype;
2512 contigfree(base, size, M_DEVBUF);
2516 MmSizeOfMdl(vaddr, len)
2522 l = sizeof(struct mdl) +
2523 (sizeof(vm_offset_t *) * SPAN_PAGES(vaddr, len));
2529 * The Microsoft documentation says this routine fills in the
2530 * page array of an MDL with the _physical_ page addresses that
2531 * comprise the buffer, but we don't really want to do that here.
2532 * Instead, we just fill in the page array with the kernel virtual
2533 * addresses of the buffers.
2536 MmBuildMdlForNonPagedPool(m)
2539 vm_offset_t *mdl_pages;
2542 pagecnt = SPAN_PAGES(m->mdl_byteoffset, m->mdl_bytecount);
2544 if (pagecnt > (m->mdl_size - sizeof(mdl)) / sizeof(vm_offset_t *))
2545 panic("not enough pages in MDL to describe buffer");
2547 mdl_pages = MmGetMdlPfnArray(m);
2549 for (i = 0; i < pagecnt; i++)
2550 *mdl_pages = (vm_offset_t)m->mdl_startva + (i * PAGE_SIZE);
2552 m->mdl_flags |= MDL_SOURCE_IS_NONPAGED_POOL;
2553 m->mdl_mappedsystemva = MmGetMdlVirtualAddress(m);
2557 MmMapLockedPages(mdl *buf, uint8_t accessmode)
2559 buf->mdl_flags |= MDL_MAPPED_TO_SYSTEM_VA;
2560 return (MmGetMdlVirtualAddress(buf));
2564 MmMapLockedPagesSpecifyCache(mdl *buf, uint8_t accessmode, uint32_t cachetype,
2565 void *vaddr, uint32_t bugcheck, uint32_t prio)
2567 return (MmMapLockedPages(buf, accessmode));
2571 MmUnmapLockedPages(vaddr, buf)
2575 buf->mdl_flags &= ~MDL_MAPPED_TO_SYSTEM_VA;
2579 * This function has a problem in that it will break if you
2580 * compile this module without PAE and try to use it on a PAE
2581 * kernel. Unfortunately, there's no way around this at the
2582 * moment. It's slightly less broken that using pmap_kextract().
2583 * You'd think the virtual memory subsystem would help us out
2584 * here, but it doesn't.
2588 MmGetPhysicalAddress(void *base)
2590 return (pmap_extract(kernel_map->pmap, (vm_offset_t)base));
2594 MmGetSystemRoutineAddress(ustr)
2595 unicode_string *ustr;
2599 if (RtlUnicodeStringToAnsiString(&astr, ustr, TRUE))
2601 return (ndis_get_routine_address(ntoskrnl_functbl, astr.as_buf));
2605 MmIsAddressValid(vaddr)
2608 if (pmap_extract(kernel_map->pmap, (vm_offset_t)vaddr))
2615 MmMapIoSpace(paddr, len, cachetype)
2620 devclass_t nexus_class;
2621 device_t *nexus_devs, devp;
2622 int nexus_count = 0;
2623 device_t matching_dev = NULL;
2624 struct resource *res;
2628 /* There will always be at least one nexus. */
2630 nexus_class = devclass_find("nexus");
2631 devclass_get_devices(nexus_class, &nexus_devs, &nexus_count);
2633 for (i = 0; i < nexus_count; i++) {
2634 devp = nexus_devs[i];
2635 matching_dev = ntoskrnl_finddev(devp, paddr, &res);
2640 free(nexus_devs, M_TEMP);
2642 if (matching_dev == NULL)
2645 v = (vm_offset_t)rman_get_virtual(res);
2646 if (paddr > rman_get_start(res))
2647 v += paddr - rman_get_start(res);
2653 MmUnmapIoSpace(vaddr, len)
2661 ntoskrnl_finddev(dev, paddr, res)
2664 struct resource **res;
2666 device_t *children = NULL;
2667 device_t matching_dev;
2670 struct resource_list *rl;
2671 struct resource_list_entry *rle;
2675 /* We only want devices that have been successfully probed. */
2677 if (device_is_alive(dev) == FALSE)
2680 rl = BUS_GET_RESOURCE_LIST(device_get_parent(dev), dev);
2682 STAILQ_FOREACH(rle, rl, link) {
2688 flags = rman_get_flags(r);
2690 if (rle->type == SYS_RES_MEMORY &&
2691 paddr >= rman_get_start(r) &&
2692 paddr <= rman_get_end(r)) {
2693 if (!(flags & RF_ACTIVE))
2694 bus_activate_resource(dev,
2695 SYS_RES_MEMORY, 0, r);
2703 * If this device has children, do another
2704 * level of recursion to inspect them.
2707 device_get_children(dev, &children, &childcnt);
2709 for (i = 0; i < childcnt; i++) {
2710 matching_dev = ntoskrnl_finddev(children[i], paddr, res);
2711 if (matching_dev != NULL) {
2712 free(children, M_TEMP);
2713 return (matching_dev);
2718 /* Won't somebody please think of the children! */
2720 if (children != NULL)
2721 free(children, M_TEMP);
2727 * Workitems are unlike DPCs, in that they run in a user-mode thread
2728 * context rather than at DISPATCH_LEVEL in kernel context. In our
2729 * case we run them in kernel context anyway.
2732 ntoskrnl_workitem_thread(arg)
2742 InitializeListHead(&kq->kq_disp);
2743 kq->kq_td = curthread;
2745 KeInitializeSpinLock(&kq->kq_lock);
2746 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
2749 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
2751 KeAcquireSpinLock(&kq->kq_lock, &irql);
2755 KeReleaseSpinLock(&kq->kq_lock, irql);
2759 while (!IsListEmpty(&kq->kq_disp)) {
2760 l = RemoveHeadList(&kq->kq_disp);
2761 iw = CONTAINING_RECORD(l,
2762 io_workitem, iw_listentry);
2763 InitializeListHead((&iw->iw_listentry));
2764 if (iw->iw_func == NULL)
2766 KeReleaseSpinLock(&kq->kq_lock, irql);
2767 MSCALL2(iw->iw_func, iw->iw_dobj, iw->iw_ctx);
2768 KeAcquireSpinLock(&kq->kq_lock, &irql);
2771 KeReleaseSpinLock(&kq->kq_lock, irql);
2775 return; /* notreached */
2779 RtlCharToInteger(src, base, val)
2788 return (STATUS_ACCESS_VIOLATION);
2789 while (*src != '\0' && *src <= ' ')
2793 else if (*src == '-') {
2804 } else if (*src == 'o') {
2807 } else if (*src == 'x') {
2812 } else if (!(base == 2 || base == 8 || base == 10 || base == 16))
2813 return (STATUS_INVALID_PARAMETER);
2815 for (res = 0; *src; src++) {
2819 else if (isxdigit(*src))
2820 v = tolower(*src) - 'a' + 10;
2824 return (STATUS_INVALID_PARAMETER);
2825 res = res * base + v;
2827 *val = negative ? -res : res;
2828 return (STATUS_SUCCESS);
2832 ntoskrnl_destroy_workitem_threads(void)
2837 for (i = 0; i < WORKITEM_THREADS; i++) {
2840 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2842 tsleep(kq->kq_td->td_proc, PWAIT, "waitiw", hz/10);
2847 IoAllocateWorkItem(dobj)
2848 device_object *dobj;
2852 iw = uma_zalloc(iw_zone, M_NOWAIT);
2856 InitializeListHead(&iw->iw_listentry);
2859 mtx_lock(&ntoskrnl_dispatchlock);
2860 iw->iw_idx = wq_idx;
2861 WORKIDX_INC(wq_idx);
2862 mtx_unlock(&ntoskrnl_dispatchlock);
2871 uma_zfree(iw_zone, iw);
2875 IoQueueWorkItem(iw, iw_func, qtype, ctx)
2877 io_workitem_func iw_func;
2886 kq = wq_queues + iw->iw_idx;
2888 KeAcquireSpinLock(&kq->kq_lock, &irql);
2891 * Traverse the list and make sure this workitem hasn't
2892 * already been inserted. Queuing the same workitem
2893 * twice will hose the list but good.
2896 l = kq->kq_disp.nle_flink;
2897 while (l != &kq->kq_disp) {
2898 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2900 /* Already queued -- do nothing. */
2901 KeReleaseSpinLock(&kq->kq_lock, irql);
2907 iw->iw_func = iw_func;
2910 InsertTailList((&kq->kq_disp), (&iw->iw_listentry));
2911 KeReleaseSpinLock(&kq->kq_lock, irql);
2913 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2917 ntoskrnl_workitem(dobj, arg)
2918 device_object *dobj;
2926 w = (work_queue_item *)dobj;
2927 f = (work_item_func)w->wqi_func;
2928 uma_zfree(iw_zone, iw);
2929 MSCALL2(f, w, w->wqi_ctx);
2933 * The ExQueueWorkItem() API is deprecated in Windows XP. Microsoft
2934 * warns that it's unsafe and to use IoQueueWorkItem() instead. The
2935 * problem with ExQueueWorkItem() is that it can't guard against
2936 * the condition where a driver submits a job to the work queue and
2937 * is then unloaded before the job is able to run. IoQueueWorkItem()
2938 * acquires a reference to the device's device_object via the
2939 * object manager and retains it until after the job has completed,
2940 * which prevents the driver from being unloaded before the job
2941 * runs. (We don't currently support this behavior, though hopefully
2942 * that will change once the object manager API is fleshed out a bit.)
2944 * Having said all that, the ExQueueWorkItem() API remains, because
2945 * there are still other parts of Windows that use it, including
2946 * NDIS itself: NdisScheduleWorkItem() calls ExQueueWorkItem().
2947 * We fake up the ExQueueWorkItem() API on top of our implementation
2948 * of IoQueueWorkItem(). Workitem thread #3 is reserved exclusively
2949 * for ExQueueWorkItem() jobs, and we pass a pointer to the work
2950 * queue item (provided by the caller) in to IoAllocateWorkItem()
2951 * instead of the device_object. We need to save this pointer so
2952 * we can apply a sanity check: as with the DPC queue and other
2953 * workitem queues, we can't allow the same work queue item to
2954 * be queued twice. If it's already pending, we silently return
2958 ExQueueWorkItem(w, qtype)
2963 io_workitem_func iwf;
2971 * We need to do a special sanity test to make sure
2972 * the ExQueueWorkItem() API isn't used to queue
2973 * the same workitem twice. Rather than checking the
2974 * io_workitem pointer itself, we test the attached
2975 * device object, which is really a pointer to the
2976 * legacy work queue item structure.
2979 kq = wq_queues + WORKITEM_LEGACY_THREAD;
2980 KeAcquireSpinLock(&kq->kq_lock, &irql);
2981 l = kq->kq_disp.nle_flink;
2982 while (l != &kq->kq_disp) {
2983 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2984 if (cur->iw_dobj == (device_object *)w) {
2985 /* Already queued -- do nothing. */
2986 KeReleaseSpinLock(&kq->kq_lock, irql);
2991 KeReleaseSpinLock(&kq->kq_lock, irql);
2993 iw = IoAllocateWorkItem((device_object *)w);
2997 iw->iw_idx = WORKITEM_LEGACY_THREAD;
2998 iwf = (io_workitem_func)ntoskrnl_findwrap((funcptr)ntoskrnl_workitem);
2999 IoQueueWorkItem(iw, iwf, qtype, iw);
3003 RtlZeroMemory(dst, len)
3011 RtlSecureZeroMemory(dst, len)
3015 memset(dst, 0, len);
3019 RtlFillMemory(void *dst, size_t len, uint8_t c)
3021 memset(dst, c, len);
3025 RtlMoveMemory(dst, src, len)
3030 memmove(dst, src, len);
3034 RtlCopyMemory(dst, src, len)
3039 bcopy(src, dst, len);
3043 RtlCompareMemory(s1, s2, len)
3051 m1 = __DECONST(char *, s1);
3052 m2 = __DECONST(char *, s2);
3054 for (i = 0; i < len && m1[i] == m2[i]; i++);
3059 RtlInitAnsiString(dst, src)
3069 a->as_len = a->as_maxlen = 0;
3073 a->as_len = a->as_maxlen = strlen(src);
3078 RtlInitUnicodeString(dst, src)
3079 unicode_string *dst;
3089 u->us_len = u->us_maxlen = 0;
3096 u->us_len = u->us_maxlen = i * 2;
3101 RtlUnicodeStringToInteger(ustr, base, val)
3102 unicode_string *ustr;
3111 uchr = ustr->us_buf;
3113 bzero(abuf, sizeof(abuf));
3115 if ((char)((*uchr) & 0xFF) == '-') {
3119 } else if ((char)((*uchr) & 0xFF) == '+') {
3126 if ((char)((*uchr) & 0xFF) == 'b') {
3130 } else if ((char)((*uchr) & 0xFF) == 'o') {
3134 } else if ((char)((*uchr) & 0xFF) == 'x') {
3148 ntoskrnl_unicode_to_ascii(uchr, astr, len);
3149 *val = strtoul(abuf, NULL, base);
3151 return (STATUS_SUCCESS);
3155 RtlFreeUnicodeString(ustr)
3156 unicode_string *ustr;
3158 if (ustr->us_buf == NULL)
3160 ExFreePool(ustr->us_buf);
3161 ustr->us_buf = NULL;
3165 RtlFreeAnsiString(astr)
3168 if (astr->as_buf == NULL)
3170 ExFreePool(astr->as_buf);
3171 astr->as_buf = NULL;
3178 return (int)strtol(str, (char **)NULL, 10);
3185 return strtol(str, (char **)NULL, 10);
3194 srandom(tv.tv_usec);
3195 return ((int)random());
3206 IoIsWdmVersionAvailable(uint8_t major, uint8_t minor)
3208 if (major == WDM_MAJOR && minor == WDM_MINOR_WINXP)
3214 IoOpenDeviceRegistryKey(struct device_object *devobj, uint32_t type,
3215 uint32_t mask, void **key)
3217 return (NDIS_STATUS_INVALID_DEVICE_REQUEST);
3221 IoGetDeviceObjectPointer(name, reqaccess, fileobj, devobj)
3222 unicode_string *name;
3225 device_object *devobj;
3227 return (STATUS_SUCCESS);
3231 IoGetDeviceProperty(devobj, regprop, buflen, prop, reslen)
3232 device_object *devobj;
3241 drv = devobj->do_drvobj;
3244 case DEVPROP_DRIVER_KEYNAME:
3246 *name = drv->dro_drivername.us_buf;
3247 *reslen = drv->dro_drivername.us_len;
3250 return (STATUS_INVALID_PARAMETER_2);
3254 return (STATUS_SUCCESS);
3258 KeInitializeMutex(kmutex, level)
3262 InitializeListHead((&kmutex->km_header.dh_waitlisthead));
3263 kmutex->km_abandoned = FALSE;
3264 kmutex->km_apcdisable = 1;
3265 kmutex->km_header.dh_sigstate = 1;
3266 kmutex->km_header.dh_type = DISP_TYPE_MUTANT;
3267 kmutex->km_header.dh_size = sizeof(kmutant) / sizeof(uint32_t);
3268 kmutex->km_ownerthread = NULL;
3272 KeReleaseMutex(kmutant *kmutex, uint8_t kwait)
3276 mtx_lock(&ntoskrnl_dispatchlock);
3277 prevstate = kmutex->km_header.dh_sigstate;
3278 if (kmutex->km_ownerthread != curthread) {
3279 mtx_unlock(&ntoskrnl_dispatchlock);
3280 return (STATUS_MUTANT_NOT_OWNED);
3283 kmutex->km_header.dh_sigstate++;
3284 kmutex->km_abandoned = FALSE;
3286 if (kmutex->km_header.dh_sigstate == 1) {
3287 kmutex->km_ownerthread = NULL;
3288 ntoskrnl_waittest(&kmutex->km_header, IO_NO_INCREMENT);
3291 mtx_unlock(&ntoskrnl_dispatchlock);
3297 KeReadStateMutex(kmutex)
3300 return (kmutex->km_header.dh_sigstate);
3304 KeInitializeEvent(nt_kevent *kevent, uint32_t type, uint8_t state)
3306 InitializeListHead((&kevent->k_header.dh_waitlisthead));
3307 kevent->k_header.dh_sigstate = state;
3308 if (type == EVENT_TYPE_NOTIFY)
3309 kevent->k_header.dh_type = DISP_TYPE_NOTIFICATION_EVENT;
3311 kevent->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_EVENT;
3312 kevent->k_header.dh_size = sizeof(nt_kevent) / sizeof(uint32_t);
3316 KeResetEvent(kevent)
3321 mtx_lock(&ntoskrnl_dispatchlock);
3322 prevstate = kevent->k_header.dh_sigstate;
3323 kevent->k_header.dh_sigstate = FALSE;
3324 mtx_unlock(&ntoskrnl_dispatchlock);
3330 KeSetEvent(nt_kevent *kevent, uint32_t increment, uint8_t kwait)
3334 nt_dispatch_header *dh;
3338 mtx_lock(&ntoskrnl_dispatchlock);
3339 prevstate = kevent->k_header.dh_sigstate;
3340 dh = &kevent->k_header;
3342 if (IsListEmpty(&dh->dh_waitlisthead))
3344 * If there's nobody in the waitlist, just set
3345 * the state to signalled.
3347 dh->dh_sigstate = 1;
3350 * Get the first waiter. If this is a synchronization
3351 * event, just wake up that one thread (don't bother
3352 * setting the state to signalled since we're supposed
3353 * to automatically clear synchronization events anyway).
3355 * If it's a notification event, or the first
3356 * waiter is doing a WAITTYPE_ALL wait, go through
3357 * the full wait satisfaction process.
3359 w = CONTAINING_RECORD(dh->dh_waitlisthead.nle_flink,
3360 wait_block, wb_waitlist);
3363 if (kevent->k_header.dh_type == DISP_TYPE_NOTIFICATION_EVENT ||
3364 w->wb_waittype == WAITTYPE_ALL) {
3365 if (prevstate == 0) {
3366 dh->dh_sigstate = 1;
3367 ntoskrnl_waittest(dh, increment);
3370 w->wb_awakened |= TRUE;
3371 cv_broadcastpri(&we->we_cv,
3372 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
3373 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
3377 mtx_unlock(&ntoskrnl_dispatchlock);
3383 KeClearEvent(kevent)
3386 kevent->k_header.dh_sigstate = FALSE;
3390 KeReadStateEvent(kevent)
3393 return (kevent->k_header.dh_sigstate);
3397 * The object manager in Windows is responsible for managing
3398 * references and access to various types of objects, including
3399 * device_objects, events, threads, timers and so on. However,
3400 * there's a difference in the way objects are handled in user
3401 * mode versus kernel mode.
3403 * In user mode (i.e. Win32 applications), all objects are
3404 * managed by the object manager. For example, when you create
3405 * a timer or event object, you actually end up with an
3406 * object_header (for the object manager's bookkeeping
3407 * purposes) and an object body (which contains the actual object
3408 * structure, e.g. ktimer, kevent, etc...). This allows Windows
3409 * to manage resource quotas and to enforce access restrictions
3410 * on basically every kind of system object handled by the kernel.
3412 * However, in kernel mode, you only end up using the object
3413 * manager some of the time. For example, in a driver, you create
3414 * a timer object by simply allocating the memory for a ktimer
3415 * structure and initializing it with KeInitializeTimer(). Hence,
3416 * the timer has no object_header and no reference counting or
3417 * security/resource checks are done on it. The assumption in
3418 * this case is that if you're running in kernel mode, you know
3419 * what you're doing, and you're already at an elevated privilege
3422 * There are some exceptions to this. The two most important ones
3423 * for our purposes are device_objects and threads. We need to use
3424 * the object manager to do reference counting on device_objects,
3425 * and for threads, you can only get a pointer to a thread's
3426 * dispatch header by using ObReferenceObjectByHandle() on the
3427 * handle returned by PsCreateSystemThread().
3431 ObReferenceObjectByHandle(ndis_handle handle, uint32_t reqaccess, void *otype,
3432 uint8_t accessmode, void **object, void **handleinfo)
3436 nr = malloc(sizeof(nt_objref), M_DEVBUF, M_NOWAIT|M_ZERO);
3438 return (STATUS_INSUFFICIENT_RESOURCES);
3440 InitializeListHead((&nr->no_dh.dh_waitlisthead));
3441 nr->no_obj = handle;
3442 nr->no_dh.dh_type = DISP_TYPE_THREAD;
3443 nr->no_dh.dh_sigstate = 0;
3444 nr->no_dh.dh_size = (uint8_t)(sizeof(struct thread) /
3446 TAILQ_INSERT_TAIL(&ntoskrnl_reflist, nr, link);
3449 return (STATUS_SUCCESS);
3453 ObfDereferenceObject(object)
3459 TAILQ_REMOVE(&ntoskrnl_reflist, nr, link);
3467 return (STATUS_SUCCESS);
3471 WmiQueryTraceInformation(traceclass, traceinfo, infolen, reqlen, buf)
3472 uint32_t traceclass;
3478 return (STATUS_NOT_FOUND);
3482 WmiTraceMessage(uint64_t loghandle, uint32_t messageflags,
3483 void *guid, uint16_t messagenum, ...)
3485 return (STATUS_SUCCESS);
3489 IoWMIRegistrationControl(dobj, action)
3490 device_object *dobj;
3493 return (STATUS_SUCCESS);
3497 * This is here just in case the thread returns without calling
3498 * PsTerminateSystemThread().
3501 ntoskrnl_thrfunc(arg)
3504 thread_context *thrctx;
3505 uint32_t (*tfunc)(void *);
3510 tfunc = thrctx->tc_thrfunc;
3511 tctx = thrctx->tc_thrctx;
3512 free(thrctx, M_TEMP);
3514 rval = MSCALL1(tfunc, tctx);
3516 PsTerminateSystemThread(rval);
3517 return; /* notreached */
3521 PsCreateSystemThread(handle, reqaccess, objattrs, phandle,
3522 clientid, thrfunc, thrctx)
3523 ndis_handle *handle;
3526 ndis_handle phandle;
3535 tc = malloc(sizeof(thread_context), M_TEMP, M_NOWAIT);
3537 return (STATUS_INSUFFICIENT_RESOURCES);
3539 tc->tc_thrctx = thrctx;
3540 tc->tc_thrfunc = thrfunc;
3542 error = kproc_create(ntoskrnl_thrfunc, tc, &p,
3543 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Kthread %d", ntoskrnl_kth);
3547 return (STATUS_INSUFFICIENT_RESOURCES);
3553 return (STATUS_SUCCESS);
3557 * In Windows, the exit of a thread is an event that you're allowed
3558 * to wait on, assuming you've obtained a reference to the thread using
3559 * ObReferenceObjectByHandle(). Unfortunately, the only way we can
3560 * simulate this behavior is to register each thread we create in a
3561 * reference list, and if someone holds a reference to us, we poke
3565 PsTerminateSystemThread(status)
3568 struct nt_objref *nr;
3570 mtx_lock(&ntoskrnl_dispatchlock);
3571 TAILQ_FOREACH(nr, &ntoskrnl_reflist, link) {
3572 if (nr->no_obj != curthread->td_proc)
3574 nr->no_dh.dh_sigstate = 1;
3575 ntoskrnl_waittest(&nr->no_dh, IO_NO_INCREMENT);
3578 mtx_unlock(&ntoskrnl_dispatchlock);
3583 return (0); /* notreached */
3587 DbgPrint(char *fmt, ...)
3597 return (STATUS_SUCCESS);
3604 kdb_enter(KDB_WHY_NDIS, "DbgBreakPoint(): breakpoint");
3608 KeBugCheckEx(code, param1, param2, param3, param4)
3615 panic("KeBugCheckEx: STOP 0x%X", code);
3619 ntoskrnl_timercall(arg)
3626 mtx_lock(&ntoskrnl_dispatchlock);
3630 #ifdef NTOSKRNL_DEBUG_TIMERS
3631 ntoskrnl_timer_fires++;
3633 ntoskrnl_remove_timer(timer);
3636 * This should never happen, but complain
3640 if (timer->k_header.dh_inserted == FALSE) {
3641 mtx_unlock(&ntoskrnl_dispatchlock);
3642 printf("NTOS: timer %p fired even though "
3643 "it was canceled\n", timer);
3647 /* Mark the timer as no longer being on the timer queue. */
3649 timer->k_header.dh_inserted = FALSE;
3651 /* Now signal the object and satisfy any waits on it. */
3653 timer->k_header.dh_sigstate = 1;
3654 ntoskrnl_waittest(&timer->k_header, IO_NO_INCREMENT);
3657 * If this is a periodic timer, re-arm it
3658 * so it will fire again. We do this before
3659 * calling any deferred procedure calls because
3660 * it's possible the DPC might cancel the timer,
3661 * in which case it would be wrong for us to
3662 * re-arm it again afterwards.
3665 if (timer->k_period) {
3667 tv.tv_usec = timer->k_period * 1000;
3668 timer->k_header.dh_inserted = TRUE;
3669 ntoskrnl_insert_timer(timer, tvtohz(&tv));
3670 #ifdef NTOSKRNL_DEBUG_TIMERS
3671 ntoskrnl_timer_reloads++;
3677 mtx_unlock(&ntoskrnl_dispatchlock);
3679 /* If there's a DPC associated with the timer, queue it up. */
3682 KeInsertQueueDpc(dpc, NULL, NULL);
3685 #ifdef NTOSKRNL_DEBUG_TIMERS
3687 sysctl_show_timers(SYSCTL_HANDLER_ARGS)
3692 ntoskrnl_show_timers();
3693 return (sysctl_handle_int(oidp, &ret, 0, req));
3697 ntoskrnl_show_timers()
3702 mtx_lock_spin(&ntoskrnl_calllock);
3703 l = ntoskrnl_calllist.nle_flink;
3704 while(l != &ntoskrnl_calllist) {
3708 mtx_unlock_spin(&ntoskrnl_calllock);
3711 printf("%d timers available (out of %d)\n", i, NTOSKRNL_TIMEOUTS);
3712 printf("timer sets: %qu\n", ntoskrnl_timer_sets);
3713 printf("timer reloads: %qu\n", ntoskrnl_timer_reloads);
3714 printf("timer cancels: %qu\n", ntoskrnl_timer_cancels);
3715 printf("timer fires: %qu\n", ntoskrnl_timer_fires);
3721 * Must be called with dispatcher lock held.
3725 ntoskrnl_insert_timer(timer, ticks)
3734 * Try and allocate a timer.
3736 mtx_lock_spin(&ntoskrnl_calllock);
3737 if (IsListEmpty(&ntoskrnl_calllist)) {
3738 mtx_unlock_spin(&ntoskrnl_calllock);
3739 #ifdef NTOSKRNL_DEBUG_TIMERS
3740 ntoskrnl_show_timers();
3742 panic("out of timers!");
3744 l = RemoveHeadList(&ntoskrnl_calllist);
3745 mtx_unlock_spin(&ntoskrnl_calllock);
3747 e = CONTAINING_RECORD(l, callout_entry, ce_list);
3750 timer->k_callout = c;
3752 callout_init(c, CALLOUT_MPSAFE);
3753 callout_reset(c, ticks, ntoskrnl_timercall, timer);
3757 ntoskrnl_remove_timer(timer)
3762 e = (callout_entry *)timer->k_callout;
3763 callout_stop(timer->k_callout);
3765 mtx_lock_spin(&ntoskrnl_calllock);
3766 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
3767 mtx_unlock_spin(&ntoskrnl_calllock);
3771 KeInitializeTimer(timer)
3777 KeInitializeTimerEx(timer, EVENT_TYPE_NOTIFY);
3781 KeInitializeTimerEx(timer, type)
3788 bzero((char *)timer, sizeof(ktimer));
3789 InitializeListHead((&timer->k_header.dh_waitlisthead));
3790 timer->k_header.dh_sigstate = FALSE;
3791 timer->k_header.dh_inserted = FALSE;
3792 if (type == EVENT_TYPE_NOTIFY)
3793 timer->k_header.dh_type = DISP_TYPE_NOTIFICATION_TIMER;
3795 timer->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_TIMER;
3796 timer->k_header.dh_size = sizeof(ktimer) / sizeof(uint32_t);
3800 * DPC subsystem. A Windows Defered Procedure Call has the following
3802 * - It runs at DISPATCH_LEVEL.
3803 * - It can have one of 3 importance values that control when it
3804 * runs relative to other DPCs in the queue.
3805 * - On SMP systems, it can be set to run on a specific processor.
3806 * In order to satisfy the last property, we create a DPC thread for
3807 * each CPU in the system and bind it to that CPU. Each thread
3808 * maintains three queues with different importance levels, which
3809 * will be processed in order from lowest to highest.
3811 * In Windows, interrupt handlers run as DPCs. (Not to be confused
3812 * with ISRs, which run in interrupt context and can preempt DPCs.)
3813 * ISRs are given the highest importance so that they'll take
3814 * precedence over timers and other things.
3818 ntoskrnl_dpc_thread(arg)
3828 InitializeListHead(&kq->kq_disp);
3829 kq->kq_td = curthread;
3831 kq->kq_running = FALSE;
3832 KeInitializeSpinLock(&kq->kq_lock);
3833 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
3834 KeInitializeEvent(&kq->kq_done, EVENT_TYPE_SYNC, FALSE);
3837 * Elevate our priority. DPCs are used to run interrupt
3838 * handlers, and they should trigger as soon as possible
3839 * once scheduled by an ISR.
3842 thread_lock(curthread);
3843 #ifdef NTOSKRNL_MULTIPLE_DPCS
3844 sched_bind(curthread, kq->kq_cpu);
3846 sched_prio(curthread, PRI_MIN_KERN);
3847 thread_unlock(curthread);
3850 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
3852 KeAcquireSpinLock(&kq->kq_lock, &irql);
3856 KeReleaseSpinLock(&kq->kq_lock, irql);
3860 kq->kq_running = TRUE;
3862 while (!IsListEmpty(&kq->kq_disp)) {
3863 l = RemoveHeadList((&kq->kq_disp));
3864 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3865 InitializeListHead((&d->k_dpclistentry));
3866 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
3867 MSCALL4(d->k_deferedfunc, d, d->k_deferredctx,
3868 d->k_sysarg1, d->k_sysarg2);
3869 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3872 kq->kq_running = FALSE;
3874 KeReleaseSpinLock(&kq->kq_lock, irql);
3876 KeSetEvent(&kq->kq_done, IO_NO_INCREMENT, FALSE);
3880 return; /* notreached */
3884 ntoskrnl_destroy_dpc_threads(void)
3891 #ifdef NTOSKRNL_MULTIPLE_DPCS
3892 for (i = 0; i < mp_ncpus; i++) {
3894 for (i = 0; i < 1; i++) {
3899 KeInitializeDpc(&dpc, NULL, NULL);
3900 KeSetTargetProcessorDpc(&dpc, i);
3901 KeInsertQueueDpc(&dpc, NULL, NULL);
3903 tsleep(kq->kq_td->td_proc, PWAIT, "dpcw", hz/10);
3908 ntoskrnl_insert_dpc(head, dpc)
3915 l = head->nle_flink;
3917 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3923 if (dpc->k_importance == KDPC_IMPORTANCE_LOW)
3924 InsertTailList((head), (&dpc->k_dpclistentry));
3926 InsertHeadList((head), (&dpc->k_dpclistentry));
3932 KeInitializeDpc(dpc, dpcfunc, dpcctx)
3941 dpc->k_deferedfunc = dpcfunc;
3942 dpc->k_deferredctx = dpcctx;
3943 dpc->k_num = KDPC_CPU_DEFAULT;
3944 dpc->k_importance = KDPC_IMPORTANCE_MEDIUM;
3945 InitializeListHead((&dpc->k_dpclistentry));
3949 KeInsertQueueDpc(dpc, sysarg1, sysarg2)
3963 #ifdef NTOSKRNL_MULTIPLE_DPCS
3964 KeRaiseIrql(DISPATCH_LEVEL, &irql);
3967 * By default, the DPC is queued to run on the same CPU
3968 * that scheduled it.
3971 if (dpc->k_num == KDPC_CPU_DEFAULT)
3972 kq += curthread->td_oncpu;
3975 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3977 KeAcquireSpinLock(&kq->kq_lock, &irql);
3980 r = ntoskrnl_insert_dpc(&kq->kq_disp, dpc);
3982 dpc->k_sysarg1 = sysarg1;
3983 dpc->k_sysarg2 = sysarg2;
3985 KeReleaseSpinLock(&kq->kq_lock, irql);
3990 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
3996 KeRemoveQueueDpc(dpc)
4005 #ifdef NTOSKRNL_MULTIPLE_DPCS
4006 KeRaiseIrql(DISPATCH_LEVEL, &irql);
4008 kq = kq_queues + dpc->k_num;
4010 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
4013 KeAcquireSpinLock(&kq->kq_lock, &irql);
4016 if (dpc->k_dpclistentry.nle_flink == &dpc->k_dpclistentry) {
4017 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
4022 RemoveEntryList((&dpc->k_dpclistentry));
4023 InitializeListHead((&dpc->k_dpclistentry));
4025 KeReleaseSpinLock(&kq->kq_lock, irql);
4031 KeSetImportanceDpc(dpc, imp)
4035 if (imp != KDPC_IMPORTANCE_LOW &&
4036 imp != KDPC_IMPORTANCE_MEDIUM &&
4037 imp != KDPC_IMPORTANCE_HIGH)
4040 dpc->k_importance = (uint8_t)imp;
4044 KeSetTargetProcessorDpc(kdpc *dpc, uint8_t cpu)
4053 KeFlushQueuedDpcs(void)
4059 * Poke each DPC queue and wait
4060 * for them to drain.
4063 #ifdef NTOSKRNL_MULTIPLE_DPCS
4064 for (i = 0; i < mp_ncpus; i++) {
4066 for (i = 0; i < 1; i++) {
4069 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
4070 KeWaitForSingleObject(&kq->kq_done, 0, 0, TRUE, NULL);
4075 KeGetCurrentProcessorNumber(void)
4077 return ((uint32_t)curthread->td_oncpu);
4081 KeSetTimerEx(timer, duetime, period, dpc)
4094 mtx_lock(&ntoskrnl_dispatchlock);
4096 if (timer->k_header.dh_inserted == TRUE) {
4097 ntoskrnl_remove_timer(timer);
4098 #ifdef NTOSKRNL_DEBUG_TIMERS
4099 ntoskrnl_timer_cancels++;
4101 timer->k_header.dh_inserted = FALSE;
4106 timer->k_duetime = duetime;
4107 timer->k_period = period;
4108 timer->k_header.dh_sigstate = FALSE;
4112 tv.tv_sec = - (duetime) / 10000000;
4113 tv.tv_usec = (- (duetime) / 10) -
4114 (tv.tv_sec * 1000000);
4116 ntoskrnl_time(&curtime);
4117 if (duetime < curtime)
4118 tv.tv_sec = tv.tv_usec = 0;
4120 tv.tv_sec = ((duetime) - curtime) / 10000000;
4121 tv.tv_usec = ((duetime) - curtime) / 10 -
4122 (tv.tv_sec * 1000000);
4126 timer->k_header.dh_inserted = TRUE;
4127 ntoskrnl_insert_timer(timer, tvtohz(&tv));
4128 #ifdef NTOSKRNL_DEBUG_TIMERS
4129 ntoskrnl_timer_sets++;
4132 mtx_unlock(&ntoskrnl_dispatchlock);
4138 KeSetTimer(timer, duetime, dpc)
4143 return (KeSetTimerEx(timer, duetime, 0, dpc));
4147 * The Windows DDK documentation seems to say that cancelling
4148 * a timer that has a DPC will result in the DPC also being
4149 * cancelled, but this isn't really the case.
4153 KeCancelTimer(timer)
4161 mtx_lock(&ntoskrnl_dispatchlock);
4163 pending = timer->k_header.dh_inserted;
4165 if (timer->k_header.dh_inserted == TRUE) {
4166 timer->k_header.dh_inserted = FALSE;
4167 ntoskrnl_remove_timer(timer);
4168 #ifdef NTOSKRNL_DEBUG_TIMERS
4169 ntoskrnl_timer_cancels++;
4173 mtx_unlock(&ntoskrnl_dispatchlock);
4179 KeReadStateTimer(timer)
4182 return (timer->k_header.dh_sigstate);
4186 KeDelayExecutionThread(uint8_t wait_mode, uint8_t alertable, int64_t *interval)
4191 panic("invalid wait_mode %d", wait_mode);
4193 KeInitializeTimer(&timer);
4194 KeSetTimer(&timer, *interval, NULL);
4195 KeWaitForSingleObject(&timer, 0, 0, alertable, NULL);
4197 return STATUS_SUCCESS;
4201 KeQueryInterruptTime(void)
4206 getmicrouptime(&tv);
4208 ticks = tvtohz(&tv);
4210 return ticks * ((10000000 + hz - 1) / hz);
4213 static struct thread *
4214 KeGetCurrentThread(void)
4221 KeSetPriorityThread(td, pri)
4228 return LOW_REALTIME_PRIORITY;
4230 if (td->td_priority <= PRI_MIN_KERN)
4231 old = HIGH_PRIORITY;
4232 else if (td->td_priority >= PRI_MAX_KERN)
4235 old = LOW_REALTIME_PRIORITY;
4238 if (pri == HIGH_PRIORITY)
4239 sched_prio(td, PRI_MIN_KERN);
4240 if (pri == LOW_REALTIME_PRIORITY)
4241 sched_prio(td, PRI_MIN_KERN + (PRI_MAX_KERN - PRI_MIN_KERN) / 2);
4242 if (pri == LOW_PRIORITY)
4243 sched_prio(td, PRI_MAX_KERN);
4252 printf("ntoskrnl dummy called...\n");
4256 image_patch_table ntoskrnl_functbl[] = {
4257 IMPORT_SFUNC(RtlZeroMemory, 2),
4258 IMPORT_SFUNC(RtlSecureZeroMemory, 2),
4259 IMPORT_SFUNC(RtlFillMemory, 3),
4260 IMPORT_SFUNC(RtlMoveMemory, 3),
4261 IMPORT_SFUNC(RtlCharToInteger, 3),
4262 IMPORT_SFUNC(RtlCopyMemory, 3),
4263 IMPORT_SFUNC(RtlCopyString, 2),
4264 IMPORT_SFUNC(RtlCompareMemory, 3),
4265 IMPORT_SFUNC(RtlEqualUnicodeString, 3),
4266 IMPORT_SFUNC(RtlCopyUnicodeString, 2),
4267 IMPORT_SFUNC(RtlUnicodeStringToAnsiString, 3),
4268 IMPORT_SFUNC(RtlAnsiStringToUnicodeString, 3),
4269 IMPORT_SFUNC(RtlInitAnsiString, 2),
4270 IMPORT_SFUNC_MAP(RtlInitString, RtlInitAnsiString, 2),
4271 IMPORT_SFUNC(RtlInitUnicodeString, 2),
4272 IMPORT_SFUNC(RtlFreeAnsiString, 1),
4273 IMPORT_SFUNC(RtlFreeUnicodeString, 1),
4274 IMPORT_SFUNC(RtlUnicodeStringToInteger, 3),
4275 IMPORT_CFUNC(sprintf, 0),
4276 IMPORT_CFUNC(vsprintf, 0),
4277 IMPORT_CFUNC_MAP(_snprintf, snprintf, 0),
4278 IMPORT_CFUNC_MAP(_vsnprintf, vsnprintf, 0),
4279 IMPORT_CFUNC(DbgPrint, 0),
4280 IMPORT_SFUNC(DbgBreakPoint, 0),
4281 IMPORT_SFUNC(KeBugCheckEx, 5),
4282 IMPORT_CFUNC(strncmp, 0),
4283 IMPORT_CFUNC(strcmp, 0),
4284 IMPORT_CFUNC_MAP(stricmp, strcasecmp, 0),
4285 IMPORT_CFUNC(strncpy, 0),
4286 IMPORT_CFUNC(strcpy, 0),
4287 IMPORT_CFUNC(strlen, 0),
4288 IMPORT_CFUNC_MAP(toupper, ntoskrnl_toupper, 0),
4289 IMPORT_CFUNC_MAP(tolower, ntoskrnl_tolower, 0),
4290 IMPORT_CFUNC_MAP(strstr, ntoskrnl_strstr, 0),
4291 IMPORT_CFUNC_MAP(strncat, ntoskrnl_strncat, 0),
4292 IMPORT_CFUNC_MAP(strchr, index, 0),
4293 IMPORT_CFUNC_MAP(strrchr, rindex, 0),
4294 IMPORT_CFUNC(memcpy, 0),
4295 IMPORT_CFUNC_MAP(memmove, ntoskrnl_memmove, 0),
4296 IMPORT_CFUNC_MAP(memset, ntoskrnl_memset, 0),
4297 IMPORT_CFUNC_MAP(memchr, ntoskrnl_memchr, 0),
4298 IMPORT_SFUNC(IoAllocateDriverObjectExtension, 4),
4299 IMPORT_SFUNC(IoGetDriverObjectExtension, 2),
4300 IMPORT_FFUNC(IofCallDriver, 2),
4301 IMPORT_FFUNC(IofCompleteRequest, 2),
4302 IMPORT_SFUNC(IoAcquireCancelSpinLock, 1),
4303 IMPORT_SFUNC(IoReleaseCancelSpinLock, 1),
4304 IMPORT_SFUNC(IoCancelIrp, 1),
4305 IMPORT_SFUNC(IoConnectInterrupt, 11),
4306 IMPORT_SFUNC(IoDisconnectInterrupt, 1),
4307 IMPORT_SFUNC(IoCreateDevice, 7),
4308 IMPORT_SFUNC(IoDeleteDevice, 1),
4309 IMPORT_SFUNC(IoGetAttachedDevice, 1),
4310 IMPORT_SFUNC(IoAttachDeviceToDeviceStack, 2),
4311 IMPORT_SFUNC(IoDetachDevice, 1),
4312 IMPORT_SFUNC(IoBuildSynchronousFsdRequest, 7),
4313 IMPORT_SFUNC(IoBuildAsynchronousFsdRequest, 6),
4314 IMPORT_SFUNC(IoBuildDeviceIoControlRequest, 9),
4315 IMPORT_SFUNC(IoAllocateIrp, 2),
4316 IMPORT_SFUNC(IoReuseIrp, 2),
4317 IMPORT_SFUNC(IoMakeAssociatedIrp, 2),
4318 IMPORT_SFUNC(IoFreeIrp, 1),
4319 IMPORT_SFUNC(IoInitializeIrp, 3),
4320 IMPORT_SFUNC(KeAcquireInterruptSpinLock, 1),
4321 IMPORT_SFUNC(KeReleaseInterruptSpinLock, 2),
4322 IMPORT_SFUNC(KeSynchronizeExecution, 3),
4323 IMPORT_SFUNC(KeWaitForSingleObject, 5),
4324 IMPORT_SFUNC(KeWaitForMultipleObjects, 8),
4325 IMPORT_SFUNC(_allmul, 4),
4326 IMPORT_SFUNC(_alldiv, 4),
4327 IMPORT_SFUNC(_allrem, 4),
4328 IMPORT_RFUNC(_allshr, 0),
4329 IMPORT_RFUNC(_allshl, 0),
4330 IMPORT_SFUNC(_aullmul, 4),
4331 IMPORT_SFUNC(_aulldiv, 4),
4332 IMPORT_SFUNC(_aullrem, 4),
4333 IMPORT_RFUNC(_aullshr, 0),
4334 IMPORT_RFUNC(_aullshl, 0),
4335 IMPORT_CFUNC(atoi, 0),
4336 IMPORT_CFUNC(atol, 0),
4337 IMPORT_CFUNC(rand, 0),
4338 IMPORT_CFUNC(srand, 0),
4339 IMPORT_SFUNC(WRITE_REGISTER_USHORT, 2),
4340 IMPORT_SFUNC(READ_REGISTER_USHORT, 1),
4341 IMPORT_SFUNC(WRITE_REGISTER_ULONG, 2),
4342 IMPORT_SFUNC(READ_REGISTER_ULONG, 1),
4343 IMPORT_SFUNC(READ_REGISTER_UCHAR, 1),
4344 IMPORT_SFUNC(WRITE_REGISTER_UCHAR, 2),
4345 IMPORT_SFUNC(ExInitializePagedLookasideList, 7),
4346 IMPORT_SFUNC(ExDeletePagedLookasideList, 1),
4347 IMPORT_SFUNC(ExInitializeNPagedLookasideList, 7),
4348 IMPORT_SFUNC(ExDeleteNPagedLookasideList, 1),
4349 IMPORT_FFUNC(InterlockedPopEntrySList, 1),
4350 IMPORT_FFUNC(InitializeSListHead, 1),
4351 IMPORT_FFUNC(InterlockedPushEntrySList, 2),
4352 IMPORT_SFUNC(ExQueryDepthSList, 1),
4353 IMPORT_FFUNC_MAP(ExpInterlockedPopEntrySList,
4354 InterlockedPopEntrySList, 1),
4355 IMPORT_FFUNC_MAP(ExpInterlockedPushEntrySList,
4356 InterlockedPushEntrySList, 2),
4357 IMPORT_FFUNC(ExInterlockedPopEntrySList, 2),
4358 IMPORT_FFUNC(ExInterlockedPushEntrySList, 3),
4359 IMPORT_SFUNC(ExAllocatePoolWithTag, 3),
4360 IMPORT_SFUNC(ExFreePoolWithTag, 2),
4361 IMPORT_SFUNC(ExFreePool, 1),
4363 IMPORT_FFUNC(KefAcquireSpinLockAtDpcLevel, 1),
4364 IMPORT_FFUNC(KefReleaseSpinLockFromDpcLevel,1),
4365 IMPORT_FFUNC(KeAcquireSpinLockRaiseToDpc, 1),
4368 * For AMD64, we can get away with just mapping
4369 * KeAcquireSpinLockRaiseToDpc() directly to KfAcquireSpinLock()
4370 * because the calling conventions end up being the same.
4371 * On i386, we have to be careful because KfAcquireSpinLock()
4372 * is _fastcall but KeAcquireSpinLockRaiseToDpc() isn't.
4374 IMPORT_SFUNC(KeAcquireSpinLockAtDpcLevel, 1),
4375 IMPORT_SFUNC(KeReleaseSpinLockFromDpcLevel, 1),
4376 IMPORT_SFUNC_MAP(KeAcquireSpinLockRaiseToDpc, KfAcquireSpinLock, 1),
4378 IMPORT_SFUNC_MAP(KeReleaseSpinLock, KfReleaseSpinLock, 1),
4379 IMPORT_FFUNC(InterlockedIncrement, 1),
4380 IMPORT_FFUNC(InterlockedDecrement, 1),
4381 IMPORT_FFUNC(InterlockedExchange, 2),
4382 IMPORT_FFUNC(ExInterlockedAddLargeStatistic, 2),
4383 IMPORT_SFUNC(IoAllocateMdl, 5),
4384 IMPORT_SFUNC(IoFreeMdl, 1),
4385 IMPORT_SFUNC(MmAllocateContiguousMemory, 2 + 1),
4386 IMPORT_SFUNC(MmAllocateContiguousMemorySpecifyCache, 5 + 3),
4387 IMPORT_SFUNC(MmFreeContiguousMemory, 1),
4388 IMPORT_SFUNC(MmFreeContiguousMemorySpecifyCache, 3),
4389 IMPORT_SFUNC(MmSizeOfMdl, 1),
4390 IMPORT_SFUNC(MmMapLockedPages, 2),
4391 IMPORT_SFUNC(MmMapLockedPagesSpecifyCache, 6),
4392 IMPORT_SFUNC(MmUnmapLockedPages, 2),
4393 IMPORT_SFUNC(MmBuildMdlForNonPagedPool, 1),
4394 IMPORT_SFUNC(MmGetPhysicalAddress, 1),
4395 IMPORT_SFUNC(MmGetSystemRoutineAddress, 1),
4396 IMPORT_SFUNC(MmIsAddressValid, 1),
4397 IMPORT_SFUNC(MmMapIoSpace, 3 + 1),
4398 IMPORT_SFUNC(MmUnmapIoSpace, 2),
4399 IMPORT_SFUNC(KeInitializeSpinLock, 1),
4400 IMPORT_SFUNC(IoIsWdmVersionAvailable, 2),
4401 IMPORT_SFUNC(IoOpenDeviceRegistryKey, 4),
4402 IMPORT_SFUNC(IoGetDeviceObjectPointer, 4),
4403 IMPORT_SFUNC(IoGetDeviceProperty, 5),
4404 IMPORT_SFUNC(IoAllocateWorkItem, 1),
4405 IMPORT_SFUNC(IoFreeWorkItem, 1),
4406 IMPORT_SFUNC(IoQueueWorkItem, 4),
4407 IMPORT_SFUNC(ExQueueWorkItem, 2),
4408 IMPORT_SFUNC(ntoskrnl_workitem, 2),
4409 IMPORT_SFUNC(KeInitializeMutex, 2),
4410 IMPORT_SFUNC(KeReleaseMutex, 2),
4411 IMPORT_SFUNC(KeReadStateMutex, 1),
4412 IMPORT_SFUNC(KeInitializeEvent, 3),
4413 IMPORT_SFUNC(KeSetEvent, 3),
4414 IMPORT_SFUNC(KeResetEvent, 1),
4415 IMPORT_SFUNC(KeClearEvent, 1),
4416 IMPORT_SFUNC(KeReadStateEvent, 1),
4417 IMPORT_SFUNC(KeInitializeTimer, 1),
4418 IMPORT_SFUNC(KeInitializeTimerEx, 2),
4419 IMPORT_SFUNC(KeSetTimer, 3),
4420 IMPORT_SFUNC(KeSetTimerEx, 4),
4421 IMPORT_SFUNC(KeCancelTimer, 1),
4422 IMPORT_SFUNC(KeReadStateTimer, 1),
4423 IMPORT_SFUNC(KeInitializeDpc, 3),
4424 IMPORT_SFUNC(KeInsertQueueDpc, 3),
4425 IMPORT_SFUNC(KeRemoveQueueDpc, 1),
4426 IMPORT_SFUNC(KeSetImportanceDpc, 2),
4427 IMPORT_SFUNC(KeSetTargetProcessorDpc, 2),
4428 IMPORT_SFUNC(KeFlushQueuedDpcs, 0),
4429 IMPORT_SFUNC(KeGetCurrentProcessorNumber, 1),
4430 IMPORT_SFUNC(ObReferenceObjectByHandle, 6),
4431 IMPORT_FFUNC(ObfDereferenceObject, 1),
4432 IMPORT_SFUNC(ZwClose, 1),
4433 IMPORT_SFUNC(PsCreateSystemThread, 7),
4434 IMPORT_SFUNC(PsTerminateSystemThread, 1),
4435 IMPORT_SFUNC(IoWMIRegistrationControl, 2),
4436 IMPORT_SFUNC(WmiQueryTraceInformation, 5),
4437 IMPORT_CFUNC(WmiTraceMessage, 0),
4438 IMPORT_SFUNC(KeQuerySystemTime, 1),
4439 IMPORT_CFUNC(KeTickCount, 0),
4440 IMPORT_SFUNC(KeDelayExecutionThread, 3),
4441 IMPORT_SFUNC(KeQueryInterruptTime, 0),
4442 IMPORT_SFUNC(KeGetCurrentThread, 0),
4443 IMPORT_SFUNC(KeSetPriorityThread, 2),
4446 * This last entry is a catch-all for any function we haven't
4447 * implemented yet. The PE import list patching routine will
4448 * use it for any function that doesn't have an explicit match
4452 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
4456 { NULL, NULL, NULL }