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, CTLFLAG_RW, 0, 0,
84 sysctl_show_timers, "I", "Show ntoskrnl timer stats");
98 typedef struct kdpc_queue kdpc_queue;
102 struct thread *we_td;
105 typedef struct wb_ext wb_ext;
107 #define NTOSKRNL_TIMEOUTS 256
108 #ifdef NTOSKRNL_DEBUG_TIMERS
109 static uint64_t ntoskrnl_timer_fires;
110 static uint64_t ntoskrnl_timer_sets;
111 static uint64_t ntoskrnl_timer_reloads;
112 static uint64_t ntoskrnl_timer_cancels;
115 struct callout_entry {
116 struct callout ce_callout;
120 typedef struct callout_entry callout_entry;
122 static struct list_entry ntoskrnl_calllist;
123 static struct mtx ntoskrnl_calllock;
124 struct kuser_shared_data kuser_shared_data;
126 static struct list_entry ntoskrnl_intlist;
127 static kspin_lock ntoskrnl_intlock;
129 static uint8_t RtlEqualUnicodeString(unicode_string *,
130 unicode_string *, uint8_t);
131 static void RtlCopyString(ansi_string *, const ansi_string *);
132 static void RtlCopyUnicodeString(unicode_string *,
134 static irp *IoBuildSynchronousFsdRequest(uint32_t, device_object *,
135 void *, uint32_t, uint64_t *, nt_kevent *, io_status_block *);
136 static irp *IoBuildAsynchronousFsdRequest(uint32_t,
137 device_object *, void *, uint32_t, uint64_t *, io_status_block *);
138 static irp *IoBuildDeviceIoControlRequest(uint32_t,
139 device_object *, void *, uint32_t, void *, uint32_t,
140 uint8_t, nt_kevent *, io_status_block *);
141 static irp *IoAllocateIrp(uint8_t, uint8_t);
142 static void IoReuseIrp(irp *, uint32_t);
143 static void IoFreeIrp(irp *);
144 static void IoInitializeIrp(irp *, uint16_t, uint8_t);
145 static irp *IoMakeAssociatedIrp(irp *, uint8_t);
146 static uint32_t KeWaitForMultipleObjects(uint32_t,
147 nt_dispatch_header **, uint32_t, uint32_t, uint32_t, uint8_t,
148 int64_t *, wait_block *);
149 static void ntoskrnl_waittest(nt_dispatch_header *, uint32_t);
150 static void ntoskrnl_satisfy_wait(nt_dispatch_header *, struct thread *);
151 static void ntoskrnl_satisfy_multiple_waits(wait_block *);
152 static int ntoskrnl_is_signalled(nt_dispatch_header *, struct thread *);
153 static void ntoskrnl_insert_timer(ktimer *, int);
154 static void ntoskrnl_remove_timer(ktimer *);
155 #ifdef NTOSKRNL_DEBUG_TIMERS
156 static void ntoskrnl_show_timers(void);
158 static void ntoskrnl_timercall(void *);
159 static void ntoskrnl_dpc_thread(void *);
160 static void ntoskrnl_destroy_dpc_threads(void);
161 static void ntoskrnl_destroy_workitem_threads(void);
162 static void ntoskrnl_workitem_thread(void *);
163 static void ntoskrnl_workitem(device_object *, void *);
164 static void ntoskrnl_unicode_to_ascii(uint16_t *, char *, int);
165 static void ntoskrnl_ascii_to_unicode(char *, uint16_t *, int);
166 static uint8_t ntoskrnl_insert_dpc(list_entry *, kdpc *);
167 static void WRITE_REGISTER_USHORT(uint16_t *, uint16_t);
168 static uint16_t READ_REGISTER_USHORT(uint16_t *);
169 static void WRITE_REGISTER_ULONG(uint32_t *, uint32_t);
170 static uint32_t READ_REGISTER_ULONG(uint32_t *);
171 static void WRITE_REGISTER_UCHAR(uint8_t *, uint8_t);
172 static uint8_t READ_REGISTER_UCHAR(uint8_t *);
173 static int64_t _allmul(int64_t, int64_t);
174 static int64_t _alldiv(int64_t, int64_t);
175 static int64_t _allrem(int64_t, int64_t);
176 static int64_t _allshr(int64_t, uint8_t);
177 static int64_t _allshl(int64_t, uint8_t);
178 static uint64_t _aullmul(uint64_t, uint64_t);
179 static uint64_t _aulldiv(uint64_t, uint64_t);
180 static uint64_t _aullrem(uint64_t, uint64_t);
181 static uint64_t _aullshr(uint64_t, uint8_t);
182 static uint64_t _aullshl(uint64_t, uint8_t);
183 static slist_entry *ntoskrnl_pushsl(slist_header *, slist_entry *);
184 static void InitializeSListHead(slist_header *);
185 static slist_entry *ntoskrnl_popsl(slist_header *);
186 static void ExFreePoolWithTag(void *, uint32_t);
187 static void ExInitializePagedLookasideList(paged_lookaside_list *,
188 lookaside_alloc_func *, lookaside_free_func *,
189 uint32_t, size_t, uint32_t, uint16_t);
190 static void ExDeletePagedLookasideList(paged_lookaside_list *);
191 static void ExInitializeNPagedLookasideList(npaged_lookaside_list *,
192 lookaside_alloc_func *, lookaside_free_func *,
193 uint32_t, size_t, uint32_t, uint16_t);
194 static void ExDeleteNPagedLookasideList(npaged_lookaside_list *);
196 *ExInterlockedPushEntrySList(slist_header *,
197 slist_entry *, kspin_lock *);
199 *ExInterlockedPopEntrySList(slist_header *, kspin_lock *);
200 static uint32_t InterlockedIncrement(volatile uint32_t *);
201 static uint32_t InterlockedDecrement(volatile uint32_t *);
202 static void ExInterlockedAddLargeStatistic(uint64_t *, uint32_t);
203 static void *MmAllocateContiguousMemory(uint32_t, uint64_t);
204 static void *MmAllocateContiguousMemorySpecifyCache(uint32_t,
205 uint64_t, uint64_t, uint64_t, enum nt_caching_type);
206 static void MmFreeContiguousMemory(void *);
207 static void MmFreeContiguousMemorySpecifyCache(void *, uint32_t,
208 enum nt_caching_type);
209 static uint32_t MmSizeOfMdl(void *, size_t);
210 static void *MmMapLockedPages(mdl *, uint8_t);
211 static void *MmMapLockedPagesSpecifyCache(mdl *,
212 uint8_t, uint32_t, void *, uint32_t, uint32_t);
213 static void MmUnmapLockedPages(void *, mdl *);
214 static device_t ntoskrnl_finddev(device_t, uint64_t, struct resource **);
215 static void RtlZeroMemory(void *, size_t);
216 static void RtlSecureZeroMemory(void *, size_t);
217 static void RtlFillMemory(void *, size_t, uint8_t);
218 static void RtlMoveMemory(void *, const void *, size_t);
219 static ndis_status RtlCharToInteger(const char *, uint32_t, uint32_t *);
220 static void RtlCopyMemory(void *, const void *, size_t);
221 static size_t RtlCompareMemory(const void *, const void *, size_t);
222 static ndis_status RtlUnicodeStringToInteger(unicode_string *,
223 uint32_t, uint32_t *);
224 static int atoi (const char *);
225 static long atol (const char *);
226 static int rand(void);
227 static void srand(unsigned int);
228 static void KeQuerySystemTime(uint64_t *);
229 static uint32_t KeTickCount(void);
230 static uint8_t IoIsWdmVersionAvailable(uint8_t, uint8_t);
231 static int32_t IoOpenDeviceRegistryKey(struct device_object *, uint32_t,
233 static void ntoskrnl_thrfunc(void *);
234 static ndis_status PsCreateSystemThread(ndis_handle *,
235 uint32_t, void *, ndis_handle, void *, void *, void *);
236 static ndis_status PsTerminateSystemThread(ndis_status);
237 static ndis_status IoGetDeviceObjectPointer(unicode_string *,
238 uint32_t, void *, device_object *);
239 static ndis_status IoGetDeviceProperty(device_object *, uint32_t,
240 uint32_t, void *, uint32_t *);
241 static void KeInitializeMutex(kmutant *, uint32_t);
242 static uint32_t KeReleaseMutex(kmutant *, uint8_t);
243 static uint32_t KeReadStateMutex(kmutant *);
244 static ndis_status ObReferenceObjectByHandle(ndis_handle,
245 uint32_t, void *, uint8_t, void **, void **);
246 static void ObfDereferenceObject(void *);
247 static uint32_t ZwClose(ndis_handle);
248 static uint32_t WmiQueryTraceInformation(uint32_t, void *, uint32_t,
250 static uint32_t WmiTraceMessage(uint64_t, uint32_t, void *, uint16_t, ...);
251 static uint32_t IoWMIRegistrationControl(device_object *, uint32_t);
252 static void *ntoskrnl_memset(void *, int, size_t);
253 static void *ntoskrnl_memmove(void *, void *, size_t);
254 static void *ntoskrnl_memchr(void *, unsigned char, size_t);
255 static char *ntoskrnl_strstr(char *, char *);
256 static char *ntoskrnl_strncat(char *, char *, size_t);
257 static int ntoskrnl_toupper(int);
258 static int ntoskrnl_tolower(int);
259 static funcptr ntoskrnl_findwrap(funcptr);
260 static uint32_t DbgPrint(char *, ...);
261 static void DbgBreakPoint(void);
262 static void KeBugCheckEx(uint32_t, u_long, u_long, u_long, u_long);
263 static int32_t KeDelayExecutionThread(uint8_t, uint8_t, int64_t *);
264 static int32_t KeSetPriorityThread(struct thread *, int32_t);
265 static void dummy(void);
267 static struct mtx ntoskrnl_dispatchlock;
268 static struct mtx ntoskrnl_interlock;
269 static kspin_lock ntoskrnl_cancellock;
270 static int ntoskrnl_kth = 0;
271 static struct nt_objref_head ntoskrnl_reflist;
272 static uma_zone_t mdl_zone;
273 static uma_zone_t iw_zone;
274 static struct kdpc_queue *kq_queues;
275 static struct kdpc_queue *wq_queues;
276 static int wq_idx = 0;
281 image_patch_table *patch;
288 mtx_init(&ntoskrnl_dispatchlock,
289 "ntoskrnl dispatch lock", MTX_NDIS_LOCK, MTX_DEF|MTX_RECURSE);
290 mtx_init(&ntoskrnl_interlock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
291 KeInitializeSpinLock(&ntoskrnl_cancellock);
292 KeInitializeSpinLock(&ntoskrnl_intlock);
293 TAILQ_INIT(&ntoskrnl_reflist);
295 InitializeListHead(&ntoskrnl_calllist);
296 InitializeListHead(&ntoskrnl_intlist);
297 mtx_init(&ntoskrnl_calllock, MTX_NTOSKRNL_SPIN_LOCK, NULL, MTX_SPIN);
299 kq_queues = ExAllocatePoolWithTag(NonPagedPool,
300 #ifdef NTOSKRNL_MULTIPLE_DPCS
301 sizeof(kdpc_queue) * mp_ncpus, 0);
303 sizeof(kdpc_queue), 0);
306 if (kq_queues == NULL)
309 wq_queues = ExAllocatePoolWithTag(NonPagedPool,
310 sizeof(kdpc_queue) * WORKITEM_THREADS, 0);
312 if (wq_queues == NULL)
315 #ifdef NTOSKRNL_MULTIPLE_DPCS
316 bzero((char *)kq_queues, sizeof(kdpc_queue) * mp_ncpus);
318 bzero((char *)kq_queues, sizeof(kdpc_queue));
320 bzero((char *)wq_queues, sizeof(kdpc_queue) * WORKITEM_THREADS);
323 * Launch the DPC threads.
326 #ifdef NTOSKRNL_MULTIPLE_DPCS
327 for (i = 0; i < mp_ncpus; i++) {
329 for (i = 0; i < 1; i++) {
333 error = kproc_create(ntoskrnl_dpc_thread, kq, &p,
334 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows DPC %d", i);
336 panic("failed to launch DPC thread");
340 * Launch the workitem threads.
343 for (i = 0; i < WORKITEM_THREADS; i++) {
345 error = kproc_create(ntoskrnl_workitem_thread, kq, &p,
346 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Workitem %d", i);
348 panic("failed to launch workitem thread");
351 patch = ntoskrnl_functbl;
352 while (patch->ipt_func != NULL) {
353 windrv_wrap((funcptr)patch->ipt_func,
354 (funcptr *)&patch->ipt_wrap,
355 patch->ipt_argcnt, patch->ipt_ftype);
359 for (i = 0; i < NTOSKRNL_TIMEOUTS; i++) {
360 e = ExAllocatePoolWithTag(NonPagedPool,
361 sizeof(callout_entry), 0);
363 panic("failed to allocate timeouts");
364 mtx_lock_spin(&ntoskrnl_calllock);
365 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
366 mtx_unlock_spin(&ntoskrnl_calllock);
370 * MDLs are supposed to be variable size (they describe
371 * buffers containing some number of pages, but we don't
372 * know ahead of time how many pages that will be). But
373 * always allocating them off the heap is very slow. As
374 * a compromise, we create an MDL UMA zone big enough to
375 * handle any buffer requiring up to 16 pages, and we
376 * use those for any MDLs for buffers of 16 pages or less
377 * in size. For buffers larger than that (which we assume
378 * will be few and far between, we allocate the MDLs off
382 mdl_zone = uma_zcreate("Windows MDL", MDL_ZONE_SIZE,
383 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
385 iw_zone = uma_zcreate("Windows WorkItem", sizeof(io_workitem),
386 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
394 image_patch_table *patch;
398 patch = ntoskrnl_functbl;
399 while (patch->ipt_func != NULL) {
400 windrv_unwrap(patch->ipt_wrap);
404 /* Stop the workitem queues. */
405 ntoskrnl_destroy_workitem_threads();
406 /* Stop the DPC queues. */
407 ntoskrnl_destroy_dpc_threads();
409 ExFreePool(kq_queues);
410 ExFreePool(wq_queues);
412 uma_zdestroy(mdl_zone);
413 uma_zdestroy(iw_zone);
415 mtx_lock_spin(&ntoskrnl_calllock);
416 while(!IsListEmpty(&ntoskrnl_calllist)) {
417 l = RemoveHeadList(&ntoskrnl_calllist);
418 e = CONTAINING_RECORD(l, callout_entry, ce_list);
419 mtx_unlock_spin(&ntoskrnl_calllock);
421 mtx_lock_spin(&ntoskrnl_calllock);
423 mtx_unlock_spin(&ntoskrnl_calllock);
425 mtx_destroy(&ntoskrnl_dispatchlock);
426 mtx_destroy(&ntoskrnl_interlock);
427 mtx_destroy(&ntoskrnl_calllock);
433 * We need to be able to reference this externally from the wrapper;
434 * GCC only generates a local implementation of memset.
437 ntoskrnl_memset(buf, ch, size)
442 return (memset(buf, ch, size));
446 ntoskrnl_memmove(dst, src, size)
451 bcopy(src, dst, size);
456 ntoskrnl_memchr(void *buf, unsigned char ch, size_t len)
459 unsigned char *p = buf;
464 } while (--len != 0);
470 ntoskrnl_strstr(s, find)
476 if ((c = *find++) != 0) {
480 if ((sc = *s++) == 0)
483 } while (strncmp(s, find, len) != 0);
489 /* Taken from libc */
491 ntoskrnl_strncat(dst, src, n)
503 if ((*d = *s++) == 0)
527 RtlEqualUnicodeString(unicode_string *str1, unicode_string *str2,
528 uint8_t caseinsensitive)
532 if (str1->us_len != str2->us_len)
535 for (i = 0; i < str1->us_len; i++) {
536 if (caseinsensitive == TRUE) {
537 if (toupper((char)(str1->us_buf[i] & 0xFF)) !=
538 toupper((char)(str2->us_buf[i] & 0xFF)))
541 if (str1->us_buf[i] != str2->us_buf[i])
550 RtlCopyString(dst, src)
552 const ansi_string *src;
554 if (src != NULL && src->as_buf != NULL && dst->as_buf != NULL) {
555 dst->as_len = min(src->as_len, dst->as_maxlen);
556 memcpy(dst->as_buf, src->as_buf, dst->as_len);
557 if (dst->as_len < dst->as_maxlen)
558 dst->as_buf[dst->as_len] = 0;
564 RtlCopyUnicodeString(dest, src)
565 unicode_string *dest;
569 if (dest->us_maxlen >= src->us_len)
570 dest->us_len = src->us_len;
572 dest->us_len = dest->us_maxlen;
573 memcpy(dest->us_buf, src->us_buf, dest->us_len);
577 ntoskrnl_ascii_to_unicode(ascii, unicode, len)
586 for (i = 0; i < len; i++) {
587 *ustr = (uint16_t)ascii[i];
593 ntoskrnl_unicode_to_ascii(unicode, ascii, len)
602 for (i = 0; i < len / 2; i++) {
603 *astr = (uint8_t)unicode[i];
609 RtlUnicodeStringToAnsiString(ansi_string *dest, unicode_string *src, uint8_t allocate)
611 if (dest == NULL || src == NULL)
612 return (STATUS_INVALID_PARAMETER);
614 dest->as_len = src->us_len / 2;
615 if (dest->as_maxlen < dest->as_len)
616 dest->as_len = dest->as_maxlen;
618 if (allocate == TRUE) {
619 dest->as_buf = ExAllocatePoolWithTag(NonPagedPool,
620 (src->us_len / 2) + 1, 0);
621 if (dest->as_buf == NULL)
622 return (STATUS_INSUFFICIENT_RESOURCES);
623 dest->as_len = dest->as_maxlen = src->us_len / 2;
625 dest->as_len = src->us_len / 2; /* XXX */
626 if (dest->as_maxlen < dest->as_len)
627 dest->as_len = dest->as_maxlen;
630 ntoskrnl_unicode_to_ascii(src->us_buf, dest->as_buf,
633 return (STATUS_SUCCESS);
637 RtlAnsiStringToUnicodeString(unicode_string *dest, ansi_string *src,
640 if (dest == NULL || src == NULL)
641 return (STATUS_INVALID_PARAMETER);
643 if (allocate == TRUE) {
644 dest->us_buf = ExAllocatePoolWithTag(NonPagedPool,
646 if (dest->us_buf == NULL)
647 return (STATUS_INSUFFICIENT_RESOURCES);
648 dest->us_len = dest->us_maxlen = strlen(src->as_buf) * 2;
650 dest->us_len = src->as_len * 2; /* XXX */
651 if (dest->us_maxlen < dest->us_len)
652 dest->us_len = dest->us_maxlen;
655 ntoskrnl_ascii_to_unicode(src->as_buf, dest->us_buf,
658 return (STATUS_SUCCESS);
662 ExAllocatePoolWithTag(pooltype, len, tag)
669 buf = malloc(len, M_DEVBUF, M_NOWAIT|M_ZERO);
677 ExFreePoolWithTag(buf, tag)
692 IoAllocateDriverObjectExtension(drv, clid, extlen, ext)
698 custom_extension *ce;
700 ce = ExAllocatePoolWithTag(NonPagedPool, sizeof(custom_extension)
704 return (STATUS_INSUFFICIENT_RESOURCES);
707 InsertTailList((&drv->dro_driverext->dre_usrext), (&ce->ce_list));
709 *ext = (void *)(ce + 1);
711 return (STATUS_SUCCESS);
715 IoGetDriverObjectExtension(drv, clid)
720 custom_extension *ce;
723 * Sanity check. Our dummy bus drivers don't have
724 * any driver extentions.
727 if (drv->dro_driverext == NULL)
730 e = drv->dro_driverext->dre_usrext.nle_flink;
731 while (e != &drv->dro_driverext->dre_usrext) {
732 ce = (custom_extension *)e;
733 if (ce->ce_clid == clid)
734 return ((void *)(ce + 1));
743 IoCreateDevice(driver_object *drv, uint32_t devextlen, unicode_string *devname,
744 uint32_t devtype, uint32_t devchars, uint8_t exclusive,
745 device_object **newdev)
749 dev = ExAllocatePoolWithTag(NonPagedPool, sizeof(device_object), 0);
751 return (STATUS_INSUFFICIENT_RESOURCES);
753 dev->do_type = devtype;
754 dev->do_drvobj = drv;
755 dev->do_currirp = NULL;
759 dev->do_devext = ExAllocatePoolWithTag(NonPagedPool,
762 if (dev->do_devext == NULL) {
764 return (STATUS_INSUFFICIENT_RESOURCES);
767 bzero(dev->do_devext, devextlen);
769 dev->do_devext = NULL;
771 dev->do_size = sizeof(device_object) + devextlen;
773 dev->do_attacheddev = NULL;
774 dev->do_nextdev = NULL;
775 dev->do_devtype = devtype;
776 dev->do_stacksize = 1;
777 dev->do_alignreq = 1;
778 dev->do_characteristics = devchars;
779 dev->do_iotimer = NULL;
780 KeInitializeEvent(&dev->do_devlock, EVENT_TYPE_SYNC, TRUE);
783 * Vpd is used for disk/tape devices,
784 * but we don't support those. (Yet.)
788 dev->do_devobj_ext = ExAllocatePoolWithTag(NonPagedPool,
789 sizeof(devobj_extension), 0);
791 if (dev->do_devobj_ext == NULL) {
792 if (dev->do_devext != NULL)
793 ExFreePool(dev->do_devext);
795 return (STATUS_INSUFFICIENT_RESOURCES);
798 dev->do_devobj_ext->dve_type = 0;
799 dev->do_devobj_ext->dve_size = sizeof(devobj_extension);
800 dev->do_devobj_ext->dve_devobj = dev;
803 * Attach this device to the driver object's list
804 * of devices. Note: this is not the same as attaching
805 * the device to the device stack. The driver's AddDevice
806 * routine must explicitly call IoAddDeviceToDeviceStack()
810 if (drv->dro_devobj == NULL) {
811 drv->dro_devobj = dev;
812 dev->do_nextdev = NULL;
814 dev->do_nextdev = drv->dro_devobj;
815 drv->dro_devobj = dev;
820 return (STATUS_SUCCESS);
832 if (dev->do_devobj_ext != NULL)
833 ExFreePool(dev->do_devobj_ext);
835 if (dev->do_devext != NULL)
836 ExFreePool(dev->do_devext);
838 /* Unlink the device from the driver's device list. */
840 prev = dev->do_drvobj->dro_devobj;
842 dev->do_drvobj->dro_devobj = dev->do_nextdev;
844 while (prev->do_nextdev != dev)
845 prev = prev->do_nextdev;
846 prev->do_nextdev = dev->do_nextdev;
853 IoGetAttachedDevice(dev)
863 while (d->do_attacheddev != NULL)
864 d = d->do_attacheddev;
870 IoBuildSynchronousFsdRequest(func, dobj, buf, len, off, event, status)
877 io_status_block *status;
881 ip = IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status);
884 ip->irp_usrevent = event;
890 IoBuildAsynchronousFsdRequest(func, dobj, buf, len, off, status)
896 io_status_block *status;
899 io_stack_location *sl;
901 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
905 ip->irp_usriostat = status;
906 ip->irp_tail.irp_overlay.irp_thread = NULL;
908 sl = IoGetNextIrpStackLocation(ip);
909 sl->isl_major = func;
913 sl->isl_devobj = dobj;
914 sl->isl_fileobj = NULL;
915 sl->isl_completionfunc = NULL;
917 ip->irp_userbuf = buf;
919 if (dobj->do_flags & DO_BUFFERED_IO) {
920 ip->irp_assoc.irp_sysbuf =
921 ExAllocatePoolWithTag(NonPagedPool, len, 0);
922 if (ip->irp_assoc.irp_sysbuf == NULL) {
926 bcopy(buf, ip->irp_assoc.irp_sysbuf, len);
929 if (dobj->do_flags & DO_DIRECT_IO) {
930 ip->irp_mdl = IoAllocateMdl(buf, len, FALSE, FALSE, ip);
931 if (ip->irp_mdl == NULL) {
932 if (ip->irp_assoc.irp_sysbuf != NULL)
933 ExFreePool(ip->irp_assoc.irp_sysbuf);
937 ip->irp_userbuf = NULL;
938 ip->irp_assoc.irp_sysbuf = NULL;
941 if (func == IRP_MJ_READ) {
942 sl->isl_parameters.isl_read.isl_len = len;
944 sl->isl_parameters.isl_read.isl_byteoff = *off;
946 sl->isl_parameters.isl_read.isl_byteoff = 0;
949 if (func == IRP_MJ_WRITE) {
950 sl->isl_parameters.isl_write.isl_len = len;
952 sl->isl_parameters.isl_write.isl_byteoff = *off;
954 sl->isl_parameters.isl_write.isl_byteoff = 0;
961 IoBuildDeviceIoControlRequest(uint32_t iocode, device_object *dobj, void *ibuf,
962 uint32_t ilen, void *obuf, uint32_t olen, uint8_t isinternal,
963 nt_kevent *event, io_status_block *status)
966 io_stack_location *sl;
969 ip = IoAllocateIrp(dobj->do_stacksize, TRUE);
972 ip->irp_usrevent = event;
973 ip->irp_usriostat = status;
974 ip->irp_tail.irp_overlay.irp_thread = NULL;
976 sl = IoGetNextIrpStackLocation(ip);
977 sl->isl_major = isinternal == TRUE ?
978 IRP_MJ_INTERNAL_DEVICE_CONTROL : IRP_MJ_DEVICE_CONTROL;
982 sl->isl_devobj = dobj;
983 sl->isl_fileobj = NULL;
984 sl->isl_completionfunc = NULL;
985 sl->isl_parameters.isl_ioctl.isl_iocode = iocode;
986 sl->isl_parameters.isl_ioctl.isl_ibuflen = ilen;
987 sl->isl_parameters.isl_ioctl.isl_obuflen = olen;
989 switch(IO_METHOD(iocode)) {
990 case METHOD_BUFFERED:
996 ip->irp_assoc.irp_sysbuf =
997 ExAllocatePoolWithTag(NonPagedPool, buflen, 0);
998 if (ip->irp_assoc.irp_sysbuf == NULL) {
1003 if (ilen && ibuf != NULL) {
1004 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1005 bzero((char *)ip->irp_assoc.irp_sysbuf + ilen,
1008 bzero(ip->irp_assoc.irp_sysbuf, ilen);
1009 ip->irp_userbuf = obuf;
1011 case METHOD_IN_DIRECT:
1012 case METHOD_OUT_DIRECT:
1013 if (ilen && ibuf != NULL) {
1014 ip->irp_assoc.irp_sysbuf =
1015 ExAllocatePoolWithTag(NonPagedPool, ilen, 0);
1016 if (ip->irp_assoc.irp_sysbuf == NULL) {
1020 bcopy(ibuf, ip->irp_assoc.irp_sysbuf, ilen);
1022 if (olen && obuf != NULL) {
1023 ip->irp_mdl = IoAllocateMdl(obuf, olen,
1026 * Normally we would MmProbeAndLockPages()
1027 * here, but we don't have to in our
1032 case METHOD_NEITHER:
1033 ip->irp_userbuf = obuf;
1034 sl->isl_parameters.isl_ioctl.isl_type3ibuf = ibuf;
1041 * Ideally, we should associate this IRP with the calling
1049 IoAllocateIrp(uint8_t stsize, uint8_t chargequota)
1053 i = ExAllocatePoolWithTag(NonPagedPool, IoSizeOfIrp(stsize), 0);
1057 IoInitializeIrp(i, IoSizeOfIrp(stsize), stsize);
1063 IoMakeAssociatedIrp(irp *ip, uint8_t stsize)
1067 associrp = IoAllocateIrp(stsize, FALSE);
1068 if (associrp == NULL)
1071 mtx_lock(&ntoskrnl_dispatchlock);
1072 associrp->irp_flags |= IRP_ASSOCIATED_IRP;
1073 associrp->irp_tail.irp_overlay.irp_thread =
1074 ip->irp_tail.irp_overlay.irp_thread;
1075 associrp->irp_assoc.irp_master = ip;
1076 mtx_unlock(&ntoskrnl_dispatchlock);
1089 IoInitializeIrp(irp *io, uint16_t psize, uint8_t ssize)
1091 bzero((char *)io, IoSizeOfIrp(ssize));
1092 io->irp_size = psize;
1093 io->irp_stackcnt = ssize;
1094 io->irp_currentstackloc = ssize;
1095 InitializeListHead(&io->irp_thlist);
1096 io->irp_tail.irp_overlay.irp_csl =
1097 (io_stack_location *)(io + 1) + ssize;
1101 IoReuseIrp(ip, status)
1107 allocflags = ip->irp_allocflags;
1108 IoInitializeIrp(ip, ip->irp_size, ip->irp_stackcnt);
1109 ip->irp_iostat.isb_status = status;
1110 ip->irp_allocflags = allocflags;
1114 IoAcquireCancelSpinLock(uint8_t *irql)
1116 KeAcquireSpinLock(&ntoskrnl_cancellock, irql);
1120 IoReleaseCancelSpinLock(uint8_t irql)
1122 KeReleaseSpinLock(&ntoskrnl_cancellock, irql);
1126 IoCancelIrp(irp *ip)
1131 IoAcquireCancelSpinLock(&cancelirql);
1132 cfunc = IoSetCancelRoutine(ip, NULL);
1133 ip->irp_cancel = TRUE;
1134 if (cfunc == NULL) {
1135 IoReleaseCancelSpinLock(cancelirql);
1138 ip->irp_cancelirql = cancelirql;
1139 MSCALL2(cfunc, IoGetCurrentIrpStackLocation(ip)->isl_devobj, ip);
1140 return (uint8_t)IoSetCancelValue(ip, TRUE);
1144 IofCallDriver(dobj, ip)
1145 device_object *dobj;
1148 driver_object *drvobj;
1149 io_stack_location *sl;
1151 driver_dispatch disp;
1153 drvobj = dobj->do_drvobj;
1155 if (ip->irp_currentstackloc <= 0)
1156 panic("IoCallDriver(): out of stack locations");
1158 IoSetNextIrpStackLocation(ip);
1159 sl = IoGetCurrentIrpStackLocation(ip);
1161 sl->isl_devobj = dobj;
1163 disp = drvobj->dro_dispatch[sl->isl_major];
1164 status = MSCALL2(disp, dobj, ip);
1170 IofCompleteRequest(irp *ip, uint8_t prioboost)
1173 device_object *dobj;
1174 io_stack_location *sl;
1177 KASSERT(ip->irp_iostat.isb_status != STATUS_PENDING,
1178 ("incorrect IRP(%p) status (STATUS_PENDING)", ip));
1180 sl = IoGetCurrentIrpStackLocation(ip);
1181 IoSkipCurrentIrpStackLocation(ip);
1184 if (sl->isl_ctl & SL_PENDING_RETURNED)
1185 ip->irp_pendingreturned = TRUE;
1187 if (ip->irp_currentstackloc != (ip->irp_stackcnt + 1))
1188 dobj = IoGetCurrentIrpStackLocation(ip)->isl_devobj;
1192 if (sl->isl_completionfunc != NULL &&
1193 ((ip->irp_iostat.isb_status == STATUS_SUCCESS &&
1194 sl->isl_ctl & SL_INVOKE_ON_SUCCESS) ||
1195 (ip->irp_iostat.isb_status != STATUS_SUCCESS &&
1196 sl->isl_ctl & SL_INVOKE_ON_ERROR) ||
1197 (ip->irp_cancel == TRUE &&
1198 sl->isl_ctl & SL_INVOKE_ON_CANCEL))) {
1199 cf = sl->isl_completionfunc;
1200 status = MSCALL3(cf, dobj, ip, sl->isl_completionctx);
1201 if (status == STATUS_MORE_PROCESSING_REQUIRED)
1204 if ((ip->irp_currentstackloc <= ip->irp_stackcnt) &&
1205 (ip->irp_pendingreturned == TRUE))
1206 IoMarkIrpPending(ip);
1209 /* move to the next. */
1210 IoSkipCurrentIrpStackLocation(ip);
1212 } while (ip->irp_currentstackloc <= (ip->irp_stackcnt + 1));
1214 if (ip->irp_usriostat != NULL)
1215 *ip->irp_usriostat = ip->irp_iostat;
1216 if (ip->irp_usrevent != NULL)
1217 KeSetEvent(ip->irp_usrevent, prioboost, FALSE);
1219 /* Handle any associated IRPs. */
1221 if (ip->irp_flags & IRP_ASSOCIATED_IRP) {
1222 uint32_t masterirpcnt;
1226 masterirp = ip->irp_assoc.irp_master;
1228 InterlockedDecrement(&masterirp->irp_assoc.irp_irpcnt);
1230 while ((m = ip->irp_mdl) != NULL) {
1231 ip->irp_mdl = m->mdl_next;
1235 if (masterirpcnt == 0)
1236 IoCompleteRequest(masterirp, IO_NO_INCREMENT);
1240 /* With any luck, these conditions will never arise. */
1242 if (ip->irp_flags & IRP_PAGING_IO) {
1243 if (ip->irp_mdl != NULL)
1244 IoFreeMdl(ip->irp_mdl);
1258 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1259 l = ntoskrnl_intlist.nle_flink;
1260 while (l != &ntoskrnl_intlist) {
1261 iobj = CONTAINING_RECORD(l, kinterrupt, ki_list);
1262 claimed = MSCALL2(iobj->ki_svcfunc, iobj, iobj->ki_svcctx);
1263 if (claimed == TRUE)
1267 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1271 KeAcquireInterruptSpinLock(iobj)
1275 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1280 KeReleaseInterruptSpinLock(kinterrupt *iobj, uint8_t irql)
1282 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1286 KeSynchronizeExecution(iobj, syncfunc, syncctx)
1293 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1294 MSCALL1(syncfunc, syncctx);
1295 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1301 * IoConnectInterrupt() is passed only the interrupt vector and
1302 * irql that a device wants to use, but no device-specific tag
1303 * of any kind. This conflicts rather badly with FreeBSD's
1304 * bus_setup_intr(), which needs the device_t for the device
1305 * requesting interrupt delivery. In order to bypass this
1306 * inconsistency, we implement a second level of interrupt
1307 * dispatching on top of bus_setup_intr(). All devices use
1308 * ntoskrnl_intr() as their ISR, and any device requesting
1309 * interrupts will be registered with ntoskrnl_intr()'s interrupt
1310 * dispatch list. When an interrupt arrives, we walk the list
1311 * and invoke all the registered ISRs. This effectively makes all
1312 * interrupts shared, but it's the only way to duplicate the
1313 * semantics of IoConnectInterrupt() and IoDisconnectInterrupt() properly.
1317 IoConnectInterrupt(kinterrupt **iobj, void *svcfunc, void *svcctx,
1318 kspin_lock *lock, uint32_t vector, uint8_t irql, uint8_t syncirql,
1319 uint8_t imode, uint8_t shared, uint32_t affinity, uint8_t savefloat)
1323 *iobj = ExAllocatePoolWithTag(NonPagedPool, sizeof(kinterrupt), 0);
1325 return (STATUS_INSUFFICIENT_RESOURCES);
1327 (*iobj)->ki_svcfunc = svcfunc;
1328 (*iobj)->ki_svcctx = svcctx;
1331 KeInitializeSpinLock(&(*iobj)->ki_lock_priv);
1332 (*iobj)->ki_lock = &(*iobj)->ki_lock_priv;
1334 (*iobj)->ki_lock = lock;
1336 KeAcquireSpinLock(&ntoskrnl_intlock, &curirql);
1337 InsertHeadList((&ntoskrnl_intlist), (&(*iobj)->ki_list));
1338 KeReleaseSpinLock(&ntoskrnl_intlock, curirql);
1340 return (STATUS_SUCCESS);
1344 IoDisconnectInterrupt(iobj)
1352 KeAcquireSpinLock(&ntoskrnl_intlock, &irql);
1353 RemoveEntryList((&iobj->ki_list));
1354 KeReleaseSpinLock(&ntoskrnl_intlock, irql);
1360 IoAttachDeviceToDeviceStack(src, dst)
1364 device_object *attached;
1366 mtx_lock(&ntoskrnl_dispatchlock);
1367 attached = IoGetAttachedDevice(dst);
1368 attached->do_attacheddev = src;
1369 src->do_attacheddev = NULL;
1370 src->do_stacksize = attached->do_stacksize + 1;
1371 mtx_unlock(&ntoskrnl_dispatchlock);
1377 IoDetachDevice(topdev)
1378 device_object *topdev;
1380 device_object *tail;
1382 mtx_lock(&ntoskrnl_dispatchlock);
1384 /* First, break the chain. */
1385 tail = topdev->do_attacheddev;
1387 mtx_unlock(&ntoskrnl_dispatchlock);
1390 topdev->do_attacheddev = tail->do_attacheddev;
1391 topdev->do_refcnt--;
1393 /* Now reduce the stacksize count for the takm_il objects. */
1395 tail = topdev->do_attacheddev;
1396 while (tail != NULL) {
1397 tail->do_stacksize--;
1398 tail = tail->do_attacheddev;
1401 mtx_unlock(&ntoskrnl_dispatchlock);
1405 * For the most part, an object is considered signalled if
1406 * dh_sigstate == TRUE. The exception is for mutant objects
1407 * (mutexes), where the logic works like this:
1409 * - If the thread already owns the object and sigstate is
1410 * less than or equal to 0, then the object is considered
1411 * signalled (recursive acquisition).
1412 * - If dh_sigstate == 1, the object is also considered
1417 ntoskrnl_is_signalled(obj, td)
1418 nt_dispatch_header *obj;
1423 if (obj->dh_type == DISP_TYPE_MUTANT) {
1424 km = (kmutant *)obj;
1425 if ((obj->dh_sigstate <= 0 && km->km_ownerthread == td) ||
1426 obj->dh_sigstate == 1)
1431 if (obj->dh_sigstate > 0)
1437 ntoskrnl_satisfy_wait(obj, td)
1438 nt_dispatch_header *obj;
1443 switch (obj->dh_type) {
1444 case DISP_TYPE_MUTANT:
1445 km = (struct kmutant *)obj;
1448 * If sigstate reaches 0, the mutex is now
1449 * non-signalled (the new thread owns it).
1451 if (obj->dh_sigstate == 0) {
1452 km->km_ownerthread = td;
1453 if (km->km_abandoned == TRUE)
1454 km->km_abandoned = FALSE;
1457 /* Synchronization objects get reset to unsignalled. */
1458 case DISP_TYPE_SYNCHRONIZATION_EVENT:
1459 case DISP_TYPE_SYNCHRONIZATION_TIMER:
1460 obj->dh_sigstate = 0;
1462 case DISP_TYPE_SEMAPHORE:
1471 ntoskrnl_satisfy_multiple_waits(wb)
1478 td = wb->wb_kthread;
1481 ntoskrnl_satisfy_wait(wb->wb_object, td);
1482 cur->wb_awakened = TRUE;
1484 } while (cur != wb);
1487 /* Always called with dispatcher lock held. */
1489 ntoskrnl_waittest(obj, increment)
1490 nt_dispatch_header *obj;
1493 wait_block *w, *next;
1500 * Once an object has been signalled, we walk its list of
1501 * wait blocks. If a wait block can be awakened, then satisfy
1502 * waits as necessary and wake the thread.
1504 * The rules work like this:
1506 * If a wait block is marked as WAITTYPE_ANY, then
1507 * we can satisfy the wait conditions on the current
1508 * object and wake the thread right away. Satisfying
1509 * the wait also has the effect of breaking us out
1510 * of the search loop.
1512 * If the object is marked as WAITTYLE_ALL, then the
1513 * wait block will be part of a circularly linked
1514 * list of wait blocks belonging to a waiting thread
1515 * that's sleeping in KeWaitForMultipleObjects(). In
1516 * order to wake the thread, all the objects in the
1517 * wait list must be in the signalled state. If they
1518 * are, we then satisfy all of them and wake the
1523 e = obj->dh_waitlisthead.nle_flink;
1525 while (e != &obj->dh_waitlisthead && obj->dh_sigstate > 0) {
1526 w = CONTAINING_RECORD(e, wait_block, wb_waitlist);
1530 if (w->wb_waittype == WAITTYPE_ANY) {
1532 * Thread can be awakened if
1533 * any wait is satisfied.
1535 ntoskrnl_satisfy_wait(obj, td);
1537 w->wb_awakened = TRUE;
1540 * Thread can only be woken up
1541 * if all waits are satisfied.
1542 * If the thread is waiting on multiple
1543 * objects, they should all be linked
1544 * through the wb_next pointers in the
1550 if (ntoskrnl_is_signalled(obj, td) == FALSE) {
1554 next = next->wb_next;
1556 ntoskrnl_satisfy_multiple_waits(w);
1559 if (satisfied == TRUE)
1560 cv_broadcastpri(&we->we_cv,
1561 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
1562 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
1569 * Return the number of 100 nanosecond intervals since
1570 * January 1, 1601. (?!?!)
1579 *tval = (uint64_t)ts.tv_nsec / 100 + (uint64_t)ts.tv_sec * 10000000 +
1580 11644473600 * 10000000; /* 100ns ticks from 1601 to 1970 */
1584 KeQuerySystemTime(current_time)
1585 uint64_t *current_time;
1587 ntoskrnl_time(current_time);
1594 getmicrouptime(&tv);
1600 * KeWaitForSingleObject() is a tricky beast, because it can be used
1601 * with several different object types: semaphores, timers, events,
1602 * mutexes and threads. Semaphores don't appear very often, but the
1603 * other object types are quite common. KeWaitForSingleObject() is
1604 * what's normally used to acquire a mutex, and it can be used to
1605 * wait for a thread termination.
1607 * The Windows NDIS API is implemented in terms of Windows kernel
1608 * primitives, and some of the object manipulation is duplicated in
1609 * NDIS. For example, NDIS has timers and events, which are actually
1610 * Windows kevents and ktimers. Now, you're supposed to only use the
1611 * NDIS variants of these objects within the confines of the NDIS API,
1612 * but there are some naughty developers out there who will use
1613 * KeWaitForSingleObject() on NDIS timer and event objects, so we
1614 * have to support that as well. Conseqently, our NDIS timer and event
1615 * code has to be closely tied into our ntoskrnl timer and event code,
1616 * just as it is in Windows.
1618 * KeWaitForSingleObject() may do different things for different kinds
1621 * - For events, we check if the event has been signalled. If the
1622 * event is already in the signalled state, we just return immediately,
1623 * otherwise we wait for it to be set to the signalled state by someone
1624 * else calling KeSetEvent(). Events can be either synchronization or
1625 * notification events.
1627 * - For timers, if the timer has already fired and the timer is in
1628 * the signalled state, we just return, otherwise we wait on the
1629 * timer. Unlike an event, timers get signalled automatically when
1630 * they expire rather than someone having to trip them manually.
1631 * Timers initialized with KeInitializeTimer() are always notification
1632 * events: KeInitializeTimerEx() lets you initialize a timer as
1633 * either a notification or synchronization event.
1635 * - For mutexes, we try to acquire the mutex and if we can't, we wait
1636 * on the mutex until it's available and then grab it. When a mutex is
1637 * released, it enters the signalled state, which wakes up one of the
1638 * threads waiting to acquire it. Mutexes are always synchronization
1641 * - For threads, the only thing we do is wait until the thread object
1642 * enters a signalled state, which occurs when the thread terminates.
1643 * Threads are always notification events.
1645 * A notification event wakes up all threads waiting on an object. A
1646 * synchronization event wakes up just one. Also, a synchronization event
1647 * is auto-clearing, which means we automatically set the event back to
1648 * the non-signalled state once the wakeup is done.
1652 KeWaitForSingleObject(void *arg, uint32_t reason, uint32_t mode,
1653 uint8_t alertable, int64_t *duetime)
1656 struct thread *td = curthread;
1661 nt_dispatch_header *obj;
1666 return (STATUS_INVALID_PARAMETER);
1668 mtx_lock(&ntoskrnl_dispatchlock);
1670 cv_init(&we.we_cv, "KeWFS");
1674 * Check to see if this object is already signalled,
1675 * and just return without waiting if it is.
1677 if (ntoskrnl_is_signalled(obj, td) == TRUE) {
1678 /* Sanity check the signal state value. */
1679 if (obj->dh_sigstate != INT32_MIN) {
1680 ntoskrnl_satisfy_wait(obj, curthread);
1681 mtx_unlock(&ntoskrnl_dispatchlock);
1682 return (STATUS_SUCCESS);
1685 * There's a limit to how many times we can
1686 * recursively acquire a mutant. If we hit
1687 * the limit, something is very wrong.
1689 if (obj->dh_type == DISP_TYPE_MUTANT) {
1690 mtx_unlock(&ntoskrnl_dispatchlock);
1691 panic("mutant limit exceeded");
1696 bzero((char *)&w, sizeof(wait_block));
1699 w.wb_waittype = WAITTYPE_ANY;
1702 w.wb_awakened = FALSE;
1703 w.wb_oldpri = td->td_priority;
1705 InsertTailList((&obj->dh_waitlisthead), (&w.wb_waitlist));
1708 * The timeout value is specified in 100 nanosecond units
1709 * and can be a positive or negative number. If it's positive,
1710 * then the duetime is absolute, and we need to convert it
1711 * to an absolute offset relative to now in order to use it.
1712 * If it's negative, then the duetime is relative and we
1713 * just have to convert the units.
1716 if (duetime != NULL) {
1718 tv.tv_sec = - (*duetime) / 10000000;
1719 tv.tv_usec = (- (*duetime) / 10) -
1720 (tv.tv_sec * 1000000);
1722 ntoskrnl_time(&curtime);
1723 if (*duetime < curtime)
1724 tv.tv_sec = tv.tv_usec = 0;
1726 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1727 tv.tv_usec = ((*duetime) - curtime) / 10 -
1728 (tv.tv_sec * 1000000);
1733 if (duetime == NULL)
1734 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1736 error = cv_timedwait(&we.we_cv,
1737 &ntoskrnl_dispatchlock, tvtohz(&tv));
1739 RemoveEntryList(&w.wb_waitlist);
1741 cv_destroy(&we.we_cv);
1743 /* We timed out. Leave the object alone and return status. */
1745 if (error == EWOULDBLOCK) {
1746 mtx_unlock(&ntoskrnl_dispatchlock);
1747 return (STATUS_TIMEOUT);
1750 mtx_unlock(&ntoskrnl_dispatchlock);
1752 return (STATUS_SUCCESS);
1754 return (KeWaitForMultipleObjects(1, &obj, WAITTYPE_ALL, reason,
1755 mode, alertable, duetime, &w));
1760 KeWaitForMultipleObjects(uint32_t cnt, nt_dispatch_header *obj[], uint32_t wtype,
1761 uint32_t reason, uint32_t mode, uint8_t alertable, int64_t *duetime,
1762 wait_block *wb_array)
1764 struct thread *td = curthread;
1765 wait_block *whead, *w;
1766 wait_block _wb_array[MAX_WAIT_OBJECTS];
1767 nt_dispatch_header *cur;
1769 int i, wcnt = 0, error = 0;
1771 struct timespec t1, t2;
1772 uint32_t status = STATUS_SUCCESS;
1775 if (cnt > MAX_WAIT_OBJECTS)
1776 return (STATUS_INVALID_PARAMETER);
1777 if (cnt > THREAD_WAIT_OBJECTS && wb_array == NULL)
1778 return (STATUS_INVALID_PARAMETER);
1780 mtx_lock(&ntoskrnl_dispatchlock);
1782 cv_init(&we.we_cv, "KeWFM");
1785 if (wb_array == NULL)
1790 bzero((char *)whead, sizeof(wait_block) * cnt);
1792 /* First pass: see if we can satisfy any waits immediately. */
1797 for (i = 0; i < cnt; i++) {
1798 InsertTailList((&obj[i]->dh_waitlisthead),
1801 w->wb_object = obj[i];
1802 w->wb_waittype = wtype;
1804 w->wb_awakened = FALSE;
1805 w->wb_oldpri = td->td_priority;
1809 if (ntoskrnl_is_signalled(obj[i], td)) {
1811 * There's a limit to how many times
1812 * we can recursively acquire a mutant.
1813 * If we hit the limit, something
1816 if (obj[i]->dh_sigstate == INT32_MIN &&
1817 obj[i]->dh_type == DISP_TYPE_MUTANT) {
1818 mtx_unlock(&ntoskrnl_dispatchlock);
1819 panic("mutant limit exceeded");
1823 * If this is a WAITTYPE_ANY wait, then
1824 * satisfy the waited object and exit
1828 if (wtype == WAITTYPE_ANY) {
1829 ntoskrnl_satisfy_wait(obj[i], td);
1830 status = STATUS_WAIT_0 + i;
1835 w->wb_object = NULL;
1836 RemoveEntryList(&w->wb_waitlist);
1842 * If this is a WAITTYPE_ALL wait and all objects are
1843 * already signalled, satisfy the waits and exit now.
1846 if (wtype == WAITTYPE_ALL && wcnt == 0) {
1847 for (i = 0; i < cnt; i++)
1848 ntoskrnl_satisfy_wait(obj[i], td);
1849 status = STATUS_SUCCESS;
1854 * Create a circular waitblock list. The waitcount
1855 * must always be non-zero when we get here.
1858 (w - 1)->wb_next = whead;
1860 /* Wait on any objects that aren't yet signalled. */
1862 /* Calculate timeout, if any. */
1864 if (duetime != NULL) {
1866 tv.tv_sec = - (*duetime) / 10000000;
1867 tv.tv_usec = (- (*duetime) / 10) -
1868 (tv.tv_sec * 1000000);
1870 ntoskrnl_time(&curtime);
1871 if (*duetime < curtime)
1872 tv.tv_sec = tv.tv_usec = 0;
1874 tv.tv_sec = ((*duetime) - curtime) / 10000000;
1875 tv.tv_usec = ((*duetime) - curtime) / 10 -
1876 (tv.tv_sec * 1000000);
1884 if (duetime == NULL)
1885 cv_wait(&we.we_cv, &ntoskrnl_dispatchlock);
1887 error = cv_timedwait(&we.we_cv,
1888 &ntoskrnl_dispatchlock, tvtohz(&tv));
1890 /* Wait with timeout expired. */
1893 status = STATUS_TIMEOUT;
1899 /* See what's been signalled. */
1904 if (ntoskrnl_is_signalled(cur, td) == TRUE ||
1905 w->wb_awakened == TRUE) {
1906 /* Sanity check the signal state value. */
1907 if (cur->dh_sigstate == INT32_MIN &&
1908 cur->dh_type == DISP_TYPE_MUTANT) {
1909 mtx_unlock(&ntoskrnl_dispatchlock);
1910 panic("mutant limit exceeded");
1913 if (wtype == WAITTYPE_ANY) {
1914 status = w->wb_waitkey &
1920 } while (w != whead);
1923 * If all objects have been signalled, or if this
1924 * is a WAITTYPE_ANY wait and we were woke up by
1925 * someone, we can bail.
1929 status = STATUS_SUCCESS;
1934 * If this is WAITTYPE_ALL wait, and there's still
1935 * objects that haven't been signalled, deduct the
1936 * time that's elapsed so far from the timeout and
1937 * wait again (or continue waiting indefinitely if
1938 * there's no timeout).
1941 if (duetime != NULL) {
1942 tv.tv_sec -= (t2.tv_sec - t1.tv_sec);
1943 tv.tv_usec -= (t2.tv_nsec - t1.tv_nsec) / 1000;
1950 cv_destroy(&we.we_cv);
1952 for (i = 0; i < cnt; i++) {
1953 if (whead[i].wb_object != NULL)
1954 RemoveEntryList(&whead[i].wb_waitlist);
1957 mtx_unlock(&ntoskrnl_dispatchlock);
1963 WRITE_REGISTER_USHORT(uint16_t *reg, uint16_t val)
1965 bus_space_write_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1969 READ_REGISTER_USHORT(reg)
1972 return (bus_space_read_2(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1976 WRITE_REGISTER_ULONG(reg, val)
1980 bus_space_write_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
1984 READ_REGISTER_ULONG(reg)
1987 return (bus_space_read_4(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1991 READ_REGISTER_UCHAR(uint8_t *reg)
1993 return (bus_space_read_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg));
1997 WRITE_REGISTER_UCHAR(uint8_t *reg, uint8_t val)
1999 bus_space_write_1(NDIS_BUS_SPACE_MEM, 0x0, (bus_size_t)reg, val);
2051 _allshl(int64_t a, uint8_t b)
2057 _aullshl(uint64_t a, uint8_t b)
2063 _allshr(int64_t a, uint8_t b)
2069 _aullshr(uint64_t a, uint8_t b)
2074 static slist_entry *
2075 ntoskrnl_pushsl(head, entry)
2079 slist_entry *oldhead;
2081 oldhead = head->slh_list.slh_next;
2082 entry->sl_next = head->slh_list.slh_next;
2083 head->slh_list.slh_next = entry;
2084 head->slh_list.slh_depth++;
2085 head->slh_list.slh_seq++;
2091 InitializeSListHead(head)
2094 memset(head, 0, sizeof(*head));
2097 static slist_entry *
2098 ntoskrnl_popsl(head)
2103 first = head->slh_list.slh_next;
2104 if (first != NULL) {
2105 head->slh_list.slh_next = first->sl_next;
2106 head->slh_list.slh_depth--;
2107 head->slh_list.slh_seq++;
2114 * We need this to make lookaside lists work for amd64.
2115 * We pass a pointer to ExAllocatePoolWithTag() the lookaside
2116 * list structure. For amd64 to work right, this has to be a
2117 * pointer to the wrapped version of the routine, not the
2118 * original. Letting the Windows driver invoke the original
2119 * function directly will result in a convention calling
2120 * mismatch and a pretty crash. On x86, this effectively
2121 * becomes a no-op since ipt_func and ipt_wrap are the same.
2125 ntoskrnl_findwrap(func)
2128 image_patch_table *patch;
2130 patch = ntoskrnl_functbl;
2131 while (patch->ipt_func != NULL) {
2132 if ((funcptr)patch->ipt_func == func)
2133 return ((funcptr)patch->ipt_wrap);
2141 ExInitializePagedLookasideList(paged_lookaside_list *lookaside,
2142 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2143 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2145 bzero((char *)lookaside, sizeof(paged_lookaside_list));
2147 if (size < sizeof(slist_entry))
2148 lookaside->nll_l.gl_size = sizeof(slist_entry);
2150 lookaside->nll_l.gl_size = size;
2151 lookaside->nll_l.gl_tag = tag;
2152 if (allocfunc == NULL)
2153 lookaside->nll_l.gl_allocfunc =
2154 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2156 lookaside->nll_l.gl_allocfunc = allocfunc;
2158 if (freefunc == NULL)
2159 lookaside->nll_l.gl_freefunc =
2160 ntoskrnl_findwrap((funcptr)ExFreePool);
2162 lookaside->nll_l.gl_freefunc = freefunc;
2165 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2168 lookaside->nll_l.gl_type = NonPagedPool;
2169 lookaside->nll_l.gl_depth = depth;
2170 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2174 ExDeletePagedLookasideList(lookaside)
2175 paged_lookaside_list *lookaside;
2178 void (*freefunc)(void *);
2180 freefunc = lookaside->nll_l.gl_freefunc;
2181 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2182 MSCALL1(freefunc, buf);
2186 ExInitializeNPagedLookasideList(npaged_lookaside_list *lookaside,
2187 lookaside_alloc_func *allocfunc, lookaside_free_func *freefunc,
2188 uint32_t flags, size_t size, uint32_t tag, uint16_t depth)
2190 bzero((char *)lookaside, sizeof(npaged_lookaside_list));
2192 if (size < sizeof(slist_entry))
2193 lookaside->nll_l.gl_size = sizeof(slist_entry);
2195 lookaside->nll_l.gl_size = size;
2196 lookaside->nll_l.gl_tag = tag;
2197 if (allocfunc == NULL)
2198 lookaside->nll_l.gl_allocfunc =
2199 ntoskrnl_findwrap((funcptr)ExAllocatePoolWithTag);
2201 lookaside->nll_l.gl_allocfunc = allocfunc;
2203 if (freefunc == NULL)
2204 lookaside->nll_l.gl_freefunc =
2205 ntoskrnl_findwrap((funcptr)ExFreePool);
2207 lookaside->nll_l.gl_freefunc = freefunc;
2210 KeInitializeSpinLock(&lookaside->nll_obsoletelock);
2213 lookaside->nll_l.gl_type = NonPagedPool;
2214 lookaside->nll_l.gl_depth = depth;
2215 lookaside->nll_l.gl_maxdepth = LOOKASIDE_DEPTH;
2219 ExDeleteNPagedLookasideList(lookaside)
2220 npaged_lookaside_list *lookaside;
2223 void (*freefunc)(void *);
2225 freefunc = lookaside->nll_l.gl_freefunc;
2226 while((buf = ntoskrnl_popsl(&lookaside->nll_l.gl_listhead)) != NULL)
2227 MSCALL1(freefunc, buf);
2231 InterlockedPushEntrySList(head, entry)
2235 slist_entry *oldhead;
2237 mtx_lock_spin(&ntoskrnl_interlock);
2238 oldhead = ntoskrnl_pushsl(head, entry);
2239 mtx_unlock_spin(&ntoskrnl_interlock);
2245 InterlockedPopEntrySList(head)
2250 mtx_lock_spin(&ntoskrnl_interlock);
2251 first = ntoskrnl_popsl(head);
2252 mtx_unlock_spin(&ntoskrnl_interlock);
2257 static slist_entry *
2258 ExInterlockedPushEntrySList(head, entry, lock)
2263 return (InterlockedPushEntrySList(head, entry));
2266 static slist_entry *
2267 ExInterlockedPopEntrySList(head, lock)
2271 return (InterlockedPopEntrySList(head));
2275 ExQueryDepthSList(head)
2280 mtx_lock_spin(&ntoskrnl_interlock);
2281 depth = head->slh_list.slh_depth;
2282 mtx_unlock_spin(&ntoskrnl_interlock);
2288 KeInitializeSpinLock(lock)
2296 KefAcquireSpinLockAtDpcLevel(lock)
2299 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2303 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0) {
2305 #ifdef NTOSKRNL_DEBUG_SPINLOCKS
2314 KefReleaseSpinLockFromDpcLevel(lock)
2317 atomic_store_rel_int((volatile u_int *)lock, 0);
2321 KeAcquireSpinLockRaiseToDpc(kspin_lock *lock)
2325 if (KeGetCurrentIrql() > DISPATCH_LEVEL)
2326 panic("IRQL_NOT_LESS_THAN_OR_EQUAL");
2328 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
2329 KeAcquireSpinLockAtDpcLevel(lock);
2335 KeAcquireSpinLockAtDpcLevel(kspin_lock *lock)
2337 while (atomic_cmpset_acq_int((volatile u_int *)lock, 0, 1) == 0)
2342 KeReleaseSpinLockFromDpcLevel(kspin_lock *lock)
2344 atomic_store_rel_int((volatile u_int *)lock, 0);
2346 #endif /* __i386__ */
2349 InterlockedExchange(dst, val)
2350 volatile uint32_t *dst;
2355 mtx_lock_spin(&ntoskrnl_interlock);
2358 mtx_unlock_spin(&ntoskrnl_interlock);
2364 InterlockedIncrement(addend)
2365 volatile uint32_t *addend;
2367 atomic_add_long((volatile u_long *)addend, 1);
2372 InterlockedDecrement(addend)
2373 volatile uint32_t *addend;
2375 atomic_subtract_long((volatile u_long *)addend, 1);
2380 ExInterlockedAddLargeStatistic(addend, inc)
2384 mtx_lock_spin(&ntoskrnl_interlock);
2386 mtx_unlock_spin(&ntoskrnl_interlock);
2390 IoAllocateMdl(void *vaddr, uint32_t len, uint8_t secondarybuf,
2391 uint8_t chargequota, irp *iopkt)
2396 if (MmSizeOfMdl(vaddr, len) > MDL_ZONE_SIZE)
2397 m = ExAllocatePoolWithTag(NonPagedPool,
2398 MmSizeOfMdl(vaddr, len), 0);
2400 m = uma_zalloc(mdl_zone, M_NOWAIT | M_ZERO);
2407 MmInitializeMdl(m, vaddr, len);
2410 * MmInitializMdl() clears the flags field, so we
2411 * have to set this here. If the MDL came from the
2412 * MDL UMA zone, tag it so we can release it to
2413 * the right place later.
2416 m->mdl_flags = MDL_ZONE_ALLOCED;
2418 if (iopkt != NULL) {
2419 if (secondarybuf == TRUE) {
2421 last = iopkt->irp_mdl;
2422 while (last->mdl_next != NULL)
2423 last = last->mdl_next;
2426 if (iopkt->irp_mdl != NULL)
2427 panic("leaking an MDL in IoAllocateMdl()");
2442 if (m->mdl_flags & MDL_ZONE_ALLOCED)
2443 uma_zfree(mdl_zone, m);
2449 MmAllocateContiguousMemory(size, highest)
2454 size_t pagelength = roundup(size, PAGE_SIZE);
2456 addr = ExAllocatePoolWithTag(NonPagedPool, pagelength, 0);
2462 MmAllocateContiguousMemorySpecifyCache(size, lowest, highest,
2463 boundary, cachetype)
2468 enum nt_caching_type cachetype;
2470 vm_memattr_t memattr;
2473 switch (cachetype) {
2475 memattr = VM_MEMATTR_UNCACHEABLE;
2477 case MmWriteCombined:
2478 memattr = VM_MEMATTR_WRITE_COMBINING;
2480 case MmNonCachedUnordered:
2481 memattr = VM_MEMATTR_UNCACHEABLE;
2484 case MmHardwareCoherentCached:
2487 memattr = VM_MEMATTR_DEFAULT;
2491 ret = (void *)kmem_alloc_contig(kernel_map, size, M_ZERO | M_NOWAIT,
2492 lowest, highest, PAGE_SIZE, boundary, memattr);
2494 malloc_type_allocated(M_DEVBUF, round_page(size));
2499 MmFreeContiguousMemory(base)
2506 MmFreeContiguousMemorySpecifyCache(base, size, cachetype)
2509 enum nt_caching_type cachetype;
2511 contigfree(base, size, M_DEVBUF);
2515 MmSizeOfMdl(vaddr, len)
2521 l = sizeof(struct mdl) +
2522 (sizeof(vm_offset_t *) * SPAN_PAGES(vaddr, len));
2528 * The Microsoft documentation says this routine fills in the
2529 * page array of an MDL with the _physical_ page addresses that
2530 * comprise the buffer, but we don't really want to do that here.
2531 * Instead, we just fill in the page array with the kernel virtual
2532 * addresses of the buffers.
2535 MmBuildMdlForNonPagedPool(m)
2538 vm_offset_t *mdl_pages;
2541 pagecnt = SPAN_PAGES(m->mdl_byteoffset, m->mdl_bytecount);
2543 if (pagecnt > (m->mdl_size - sizeof(mdl)) / sizeof(vm_offset_t *))
2544 panic("not enough pages in MDL to describe buffer");
2546 mdl_pages = MmGetMdlPfnArray(m);
2548 for (i = 0; i < pagecnt; i++)
2549 *mdl_pages = (vm_offset_t)m->mdl_startva + (i * PAGE_SIZE);
2551 m->mdl_flags |= MDL_SOURCE_IS_NONPAGED_POOL;
2552 m->mdl_mappedsystemva = MmGetMdlVirtualAddress(m);
2556 MmMapLockedPages(mdl *buf, uint8_t accessmode)
2558 buf->mdl_flags |= MDL_MAPPED_TO_SYSTEM_VA;
2559 return (MmGetMdlVirtualAddress(buf));
2563 MmMapLockedPagesSpecifyCache(mdl *buf, uint8_t accessmode, uint32_t cachetype,
2564 void *vaddr, uint32_t bugcheck, uint32_t prio)
2566 return (MmMapLockedPages(buf, accessmode));
2570 MmUnmapLockedPages(vaddr, buf)
2574 buf->mdl_flags &= ~MDL_MAPPED_TO_SYSTEM_VA;
2578 * This function has a problem in that it will break if you
2579 * compile this module without PAE and try to use it on a PAE
2580 * kernel. Unfortunately, there's no way around this at the
2581 * moment. It's slightly less broken that using pmap_kextract().
2582 * You'd think the virtual memory subsystem would help us out
2583 * here, but it doesn't.
2587 MmGetPhysicalAddress(void *base)
2589 return (pmap_extract(kernel_map->pmap, (vm_offset_t)base));
2593 MmGetSystemRoutineAddress(ustr)
2594 unicode_string *ustr;
2598 if (RtlUnicodeStringToAnsiString(&astr, ustr, TRUE))
2600 return (ndis_get_routine_address(ntoskrnl_functbl, astr.as_buf));
2604 MmIsAddressValid(vaddr)
2607 if (pmap_extract(kernel_map->pmap, (vm_offset_t)vaddr))
2614 MmMapIoSpace(paddr, len, cachetype)
2619 devclass_t nexus_class;
2620 device_t *nexus_devs, devp;
2621 int nexus_count = 0;
2622 device_t matching_dev = NULL;
2623 struct resource *res;
2627 /* There will always be at least one nexus. */
2629 nexus_class = devclass_find("nexus");
2630 devclass_get_devices(nexus_class, &nexus_devs, &nexus_count);
2632 for (i = 0; i < nexus_count; i++) {
2633 devp = nexus_devs[i];
2634 matching_dev = ntoskrnl_finddev(devp, paddr, &res);
2639 free(nexus_devs, M_TEMP);
2641 if (matching_dev == NULL)
2644 v = (vm_offset_t)rman_get_virtual(res);
2645 if (paddr > rman_get_start(res))
2646 v += paddr - rman_get_start(res);
2652 MmUnmapIoSpace(vaddr, len)
2660 ntoskrnl_finddev(dev, paddr, res)
2663 struct resource **res;
2665 device_t *children = NULL;
2666 device_t matching_dev;
2669 struct resource_list *rl;
2670 struct resource_list_entry *rle;
2674 /* We only want devices that have been successfully probed. */
2676 if (device_is_alive(dev) == FALSE)
2679 rl = BUS_GET_RESOURCE_LIST(device_get_parent(dev), dev);
2681 STAILQ_FOREACH(rle, rl, link) {
2687 flags = rman_get_flags(r);
2689 if (rle->type == SYS_RES_MEMORY &&
2690 paddr >= rman_get_start(r) &&
2691 paddr <= rman_get_end(r)) {
2692 if (!(flags & RF_ACTIVE))
2693 bus_activate_resource(dev,
2694 SYS_RES_MEMORY, 0, r);
2702 * If this device has children, do another
2703 * level of recursion to inspect them.
2706 device_get_children(dev, &children, &childcnt);
2708 for (i = 0; i < childcnt; i++) {
2709 matching_dev = ntoskrnl_finddev(children[i], paddr, res);
2710 if (matching_dev != NULL) {
2711 free(children, M_TEMP);
2712 return (matching_dev);
2717 /* Won't somebody please think of the children! */
2719 if (children != NULL)
2720 free(children, M_TEMP);
2726 * Workitems are unlike DPCs, in that they run in a user-mode thread
2727 * context rather than at DISPATCH_LEVEL in kernel context. In our
2728 * case we run them in kernel context anyway.
2731 ntoskrnl_workitem_thread(arg)
2741 InitializeListHead(&kq->kq_disp);
2742 kq->kq_td = curthread;
2744 KeInitializeSpinLock(&kq->kq_lock);
2745 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
2748 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
2750 KeAcquireSpinLock(&kq->kq_lock, &irql);
2754 KeReleaseSpinLock(&kq->kq_lock, irql);
2758 while (!IsListEmpty(&kq->kq_disp)) {
2759 l = RemoveHeadList(&kq->kq_disp);
2760 iw = CONTAINING_RECORD(l,
2761 io_workitem, iw_listentry);
2762 InitializeListHead((&iw->iw_listentry));
2763 if (iw->iw_func == NULL)
2765 KeReleaseSpinLock(&kq->kq_lock, irql);
2766 MSCALL2(iw->iw_func, iw->iw_dobj, iw->iw_ctx);
2767 KeAcquireSpinLock(&kq->kq_lock, &irql);
2770 KeReleaseSpinLock(&kq->kq_lock, irql);
2774 return; /* notreached */
2778 RtlCharToInteger(src, base, val)
2787 return (STATUS_ACCESS_VIOLATION);
2788 while (*src != '\0' && *src <= ' ')
2792 else if (*src == '-') {
2803 } else if (*src == 'o') {
2806 } else if (*src == 'x') {
2811 } else if (!(base == 2 || base == 8 || base == 10 || base == 16))
2812 return (STATUS_INVALID_PARAMETER);
2814 for (res = 0; *src; src++) {
2818 else if (isxdigit(*src))
2819 v = tolower(*src) - 'a' + 10;
2823 return (STATUS_INVALID_PARAMETER);
2824 res = res * base + v;
2826 *val = negative ? -res : res;
2827 return (STATUS_SUCCESS);
2831 ntoskrnl_destroy_workitem_threads(void)
2836 for (i = 0; i < WORKITEM_THREADS; i++) {
2839 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2841 tsleep(kq->kq_td->td_proc, PWAIT, "waitiw", hz/10);
2846 IoAllocateWorkItem(dobj)
2847 device_object *dobj;
2851 iw = uma_zalloc(iw_zone, M_NOWAIT);
2855 InitializeListHead(&iw->iw_listentry);
2858 mtx_lock(&ntoskrnl_dispatchlock);
2859 iw->iw_idx = wq_idx;
2860 WORKIDX_INC(wq_idx);
2861 mtx_unlock(&ntoskrnl_dispatchlock);
2870 uma_zfree(iw_zone, iw);
2874 IoQueueWorkItem(iw, iw_func, qtype, ctx)
2876 io_workitem_func iw_func;
2885 kq = wq_queues + iw->iw_idx;
2887 KeAcquireSpinLock(&kq->kq_lock, &irql);
2890 * Traverse the list and make sure this workitem hasn't
2891 * already been inserted. Queuing the same workitem
2892 * twice will hose the list but good.
2895 l = kq->kq_disp.nle_flink;
2896 while (l != &kq->kq_disp) {
2897 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2899 /* Already queued -- do nothing. */
2900 KeReleaseSpinLock(&kq->kq_lock, irql);
2906 iw->iw_func = iw_func;
2909 InsertTailList((&kq->kq_disp), (&iw->iw_listentry));
2910 KeReleaseSpinLock(&kq->kq_lock, irql);
2912 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
2916 ntoskrnl_workitem(dobj, arg)
2917 device_object *dobj;
2925 w = (work_queue_item *)dobj;
2926 f = (work_item_func)w->wqi_func;
2927 uma_zfree(iw_zone, iw);
2928 MSCALL2(f, w, w->wqi_ctx);
2932 * The ExQueueWorkItem() API is deprecated in Windows XP. Microsoft
2933 * warns that it's unsafe and to use IoQueueWorkItem() instead. The
2934 * problem with ExQueueWorkItem() is that it can't guard against
2935 * the condition where a driver submits a job to the work queue and
2936 * is then unloaded before the job is able to run. IoQueueWorkItem()
2937 * acquires a reference to the device's device_object via the
2938 * object manager and retains it until after the job has completed,
2939 * which prevents the driver from being unloaded before the job
2940 * runs. (We don't currently support this behavior, though hopefully
2941 * that will change once the object manager API is fleshed out a bit.)
2943 * Having said all that, the ExQueueWorkItem() API remains, because
2944 * there are still other parts of Windows that use it, including
2945 * NDIS itself: NdisScheduleWorkItem() calls ExQueueWorkItem().
2946 * We fake up the ExQueueWorkItem() API on top of our implementation
2947 * of IoQueueWorkItem(). Workitem thread #3 is reserved exclusively
2948 * for ExQueueWorkItem() jobs, and we pass a pointer to the work
2949 * queue item (provided by the caller) in to IoAllocateWorkItem()
2950 * instead of the device_object. We need to save this pointer so
2951 * we can apply a sanity check: as with the DPC queue and other
2952 * workitem queues, we can't allow the same work queue item to
2953 * be queued twice. If it's already pending, we silently return
2957 ExQueueWorkItem(w, qtype)
2962 io_workitem_func iwf;
2970 * We need to do a special sanity test to make sure
2971 * the ExQueueWorkItem() API isn't used to queue
2972 * the same workitem twice. Rather than checking the
2973 * io_workitem pointer itself, we test the attached
2974 * device object, which is really a pointer to the
2975 * legacy work queue item structure.
2978 kq = wq_queues + WORKITEM_LEGACY_THREAD;
2979 KeAcquireSpinLock(&kq->kq_lock, &irql);
2980 l = kq->kq_disp.nle_flink;
2981 while (l != &kq->kq_disp) {
2982 cur = CONTAINING_RECORD(l, io_workitem, iw_listentry);
2983 if (cur->iw_dobj == (device_object *)w) {
2984 /* Already queued -- do nothing. */
2985 KeReleaseSpinLock(&kq->kq_lock, irql);
2990 KeReleaseSpinLock(&kq->kq_lock, irql);
2992 iw = IoAllocateWorkItem((device_object *)w);
2996 iw->iw_idx = WORKITEM_LEGACY_THREAD;
2997 iwf = (io_workitem_func)ntoskrnl_findwrap((funcptr)ntoskrnl_workitem);
2998 IoQueueWorkItem(iw, iwf, qtype, iw);
3002 RtlZeroMemory(dst, len)
3010 RtlSecureZeroMemory(dst, len)
3014 memset(dst, 0, len);
3018 RtlFillMemory(dst, len, 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);
3196 srandom(tv.tv_usec);
3197 return ((int)random());
3208 IoIsWdmVersionAvailable(uint8_t major, uint8_t minor)
3210 if (major == WDM_MAJOR && minor == WDM_MINOR_WINXP)
3216 IoOpenDeviceRegistryKey(struct device_object *devobj, uint32_t type,
3217 uint32_t mask, void **key)
3219 return (NDIS_STATUS_INVALID_DEVICE_REQUEST);
3223 IoGetDeviceObjectPointer(name, reqaccess, fileobj, devobj)
3224 unicode_string *name;
3227 device_object *devobj;
3229 return (STATUS_SUCCESS);
3233 IoGetDeviceProperty(devobj, regprop, buflen, prop, reslen)
3234 device_object *devobj;
3243 drv = devobj->do_drvobj;
3246 case DEVPROP_DRIVER_KEYNAME:
3248 *name = drv->dro_drivername.us_buf;
3249 *reslen = drv->dro_drivername.us_len;
3252 return (STATUS_INVALID_PARAMETER_2);
3256 return (STATUS_SUCCESS);
3260 KeInitializeMutex(kmutex, level)
3264 InitializeListHead((&kmutex->km_header.dh_waitlisthead));
3265 kmutex->km_abandoned = FALSE;
3266 kmutex->km_apcdisable = 1;
3267 kmutex->km_header.dh_sigstate = 1;
3268 kmutex->km_header.dh_type = DISP_TYPE_MUTANT;
3269 kmutex->km_header.dh_size = sizeof(kmutant) / sizeof(uint32_t);
3270 kmutex->km_ownerthread = NULL;
3274 KeReleaseMutex(kmutant *kmutex, uint8_t kwait)
3278 mtx_lock(&ntoskrnl_dispatchlock);
3279 prevstate = kmutex->km_header.dh_sigstate;
3280 if (kmutex->km_ownerthread != curthread) {
3281 mtx_unlock(&ntoskrnl_dispatchlock);
3282 return (STATUS_MUTANT_NOT_OWNED);
3285 kmutex->km_header.dh_sigstate++;
3286 kmutex->km_abandoned = FALSE;
3288 if (kmutex->km_header.dh_sigstate == 1) {
3289 kmutex->km_ownerthread = NULL;
3290 ntoskrnl_waittest(&kmutex->km_header, IO_NO_INCREMENT);
3293 mtx_unlock(&ntoskrnl_dispatchlock);
3299 KeReadStateMutex(kmutex)
3302 return (kmutex->km_header.dh_sigstate);
3306 KeInitializeEvent(nt_kevent *kevent, uint32_t type, uint8_t state)
3308 InitializeListHead((&kevent->k_header.dh_waitlisthead));
3309 kevent->k_header.dh_sigstate = state;
3310 if (type == EVENT_TYPE_NOTIFY)
3311 kevent->k_header.dh_type = DISP_TYPE_NOTIFICATION_EVENT;
3313 kevent->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_EVENT;
3314 kevent->k_header.dh_size = sizeof(nt_kevent) / sizeof(uint32_t);
3318 KeResetEvent(kevent)
3323 mtx_lock(&ntoskrnl_dispatchlock);
3324 prevstate = kevent->k_header.dh_sigstate;
3325 kevent->k_header.dh_sigstate = FALSE;
3326 mtx_unlock(&ntoskrnl_dispatchlock);
3332 KeSetEvent(nt_kevent *kevent, uint32_t increment, uint8_t kwait)
3336 nt_dispatch_header *dh;
3340 mtx_lock(&ntoskrnl_dispatchlock);
3341 prevstate = kevent->k_header.dh_sigstate;
3342 dh = &kevent->k_header;
3344 if (IsListEmpty(&dh->dh_waitlisthead))
3346 * If there's nobody in the waitlist, just set
3347 * the state to signalled.
3349 dh->dh_sigstate = 1;
3352 * Get the first waiter. If this is a synchronization
3353 * event, just wake up that one thread (don't bother
3354 * setting the state to signalled since we're supposed
3355 * to automatically clear synchronization events anyway).
3357 * If it's a notification event, or the the first
3358 * waiter is doing a WAITTYPE_ALL wait, go through
3359 * the full wait satisfaction process.
3361 w = CONTAINING_RECORD(dh->dh_waitlisthead.nle_flink,
3362 wait_block, wb_waitlist);
3365 if (kevent->k_header.dh_type == DISP_TYPE_NOTIFICATION_EVENT ||
3366 w->wb_waittype == WAITTYPE_ALL) {
3367 if (prevstate == 0) {
3368 dh->dh_sigstate = 1;
3369 ntoskrnl_waittest(dh, increment);
3372 w->wb_awakened |= TRUE;
3373 cv_broadcastpri(&we->we_cv,
3374 (w->wb_oldpri - (increment * 4)) > PRI_MIN_KERN ?
3375 w->wb_oldpri - (increment * 4) : PRI_MIN_KERN);
3379 mtx_unlock(&ntoskrnl_dispatchlock);
3385 KeClearEvent(kevent)
3388 kevent->k_header.dh_sigstate = FALSE;
3392 KeReadStateEvent(kevent)
3395 return (kevent->k_header.dh_sigstate);
3399 * The object manager in Windows is responsible for managing
3400 * references and access to various types of objects, including
3401 * device_objects, events, threads, timers and so on. However,
3402 * there's a difference in the way objects are handled in user
3403 * mode versus kernel mode.
3405 * In user mode (i.e. Win32 applications), all objects are
3406 * managed by the object manager. For example, when you create
3407 * a timer or event object, you actually end up with an
3408 * object_header (for the object manager's bookkeeping
3409 * purposes) and an object body (which contains the actual object
3410 * structure, e.g. ktimer, kevent, etc...). This allows Windows
3411 * to manage resource quotas and to enforce access restrictions
3412 * on basically every kind of system object handled by the kernel.
3414 * However, in kernel mode, you only end up using the object
3415 * manager some of the time. For example, in a driver, you create
3416 * a timer object by simply allocating the memory for a ktimer
3417 * structure and initializing it with KeInitializeTimer(). Hence,
3418 * the timer has no object_header and no reference counting or
3419 * security/resource checks are done on it. The assumption in
3420 * this case is that if you're running in kernel mode, you know
3421 * what you're doing, and you're already at an elevated privilege
3424 * There are some exceptions to this. The two most important ones
3425 * for our purposes are device_objects and threads. We need to use
3426 * the object manager to do reference counting on device_objects,
3427 * and for threads, you can only get a pointer to a thread's
3428 * dispatch header by using ObReferenceObjectByHandle() on the
3429 * handle returned by PsCreateSystemThread().
3433 ObReferenceObjectByHandle(ndis_handle handle, uint32_t reqaccess, void *otype,
3434 uint8_t accessmode, void **object, void **handleinfo)
3438 nr = malloc(sizeof(nt_objref), M_DEVBUF, M_NOWAIT|M_ZERO);
3440 return (STATUS_INSUFFICIENT_RESOURCES);
3442 InitializeListHead((&nr->no_dh.dh_waitlisthead));
3443 nr->no_obj = handle;
3444 nr->no_dh.dh_type = DISP_TYPE_THREAD;
3445 nr->no_dh.dh_sigstate = 0;
3446 nr->no_dh.dh_size = (uint8_t)(sizeof(struct thread) /
3448 TAILQ_INSERT_TAIL(&ntoskrnl_reflist, nr, link);
3451 return (STATUS_SUCCESS);
3455 ObfDereferenceObject(object)
3461 TAILQ_REMOVE(&ntoskrnl_reflist, nr, link);
3469 return (STATUS_SUCCESS);
3473 WmiQueryTraceInformation(traceclass, traceinfo, infolen, reqlen, buf)
3474 uint32_t traceclass;
3480 return (STATUS_NOT_FOUND);
3484 WmiTraceMessage(uint64_t loghandle, uint32_t messageflags,
3485 void *guid, uint16_t messagenum, ...)
3487 return (STATUS_SUCCESS);
3491 IoWMIRegistrationControl(dobj, action)
3492 device_object *dobj;
3495 return (STATUS_SUCCESS);
3499 * This is here just in case the thread returns without calling
3500 * PsTerminateSystemThread().
3503 ntoskrnl_thrfunc(arg)
3506 thread_context *thrctx;
3507 uint32_t (*tfunc)(void *);
3512 tfunc = thrctx->tc_thrfunc;
3513 tctx = thrctx->tc_thrctx;
3514 free(thrctx, M_TEMP);
3516 rval = MSCALL1(tfunc, tctx);
3518 PsTerminateSystemThread(rval);
3519 return; /* notreached */
3523 PsCreateSystemThread(handle, reqaccess, objattrs, phandle,
3524 clientid, thrfunc, thrctx)
3525 ndis_handle *handle;
3528 ndis_handle phandle;
3537 tc = malloc(sizeof(thread_context), M_TEMP, M_NOWAIT);
3539 return (STATUS_INSUFFICIENT_RESOURCES);
3541 tc->tc_thrctx = thrctx;
3542 tc->tc_thrfunc = thrfunc;
3544 error = kproc_create(ntoskrnl_thrfunc, tc, &p,
3545 RFHIGHPID, NDIS_KSTACK_PAGES, "Windows Kthread %d", ntoskrnl_kth);
3549 return (STATUS_INSUFFICIENT_RESOURCES);
3555 return (STATUS_SUCCESS);
3559 * In Windows, the exit of a thread is an event that you're allowed
3560 * to wait on, assuming you've obtained a reference to the thread using
3561 * ObReferenceObjectByHandle(). Unfortunately, the only way we can
3562 * simulate this behavior is to register each thread we create in a
3563 * reference list, and if someone holds a reference to us, we poke
3567 PsTerminateSystemThread(status)
3570 struct nt_objref *nr;
3572 mtx_lock(&ntoskrnl_dispatchlock);
3573 TAILQ_FOREACH(nr, &ntoskrnl_reflist, link) {
3574 if (nr->no_obj != curthread->td_proc)
3576 nr->no_dh.dh_sigstate = 1;
3577 ntoskrnl_waittest(&nr->no_dh, IO_NO_INCREMENT);
3580 mtx_unlock(&ntoskrnl_dispatchlock);
3585 return (0); /* notreached */
3589 DbgPrint(char *fmt, ...)
3598 return (STATUS_SUCCESS);
3605 kdb_enter(KDB_WHY_NDIS, "DbgBreakPoint(): breakpoint");
3609 KeBugCheckEx(code, param1, param2, param3, param4)
3616 panic("KeBugCheckEx: STOP 0x%X", code);
3620 ntoskrnl_timercall(arg)
3627 mtx_lock(&ntoskrnl_dispatchlock);
3631 #ifdef NTOSKRNL_DEBUG_TIMERS
3632 ntoskrnl_timer_fires++;
3634 ntoskrnl_remove_timer(timer);
3637 * This should never happen, but complain
3641 if (timer->k_header.dh_inserted == FALSE) {
3642 mtx_unlock(&ntoskrnl_dispatchlock);
3643 printf("NTOS: timer %p fired even though "
3644 "it was canceled\n", timer);
3648 /* Mark the timer as no longer being on the timer queue. */
3650 timer->k_header.dh_inserted = FALSE;
3652 /* Now signal the object and satisfy any waits on it. */
3654 timer->k_header.dh_sigstate = 1;
3655 ntoskrnl_waittest(&timer->k_header, IO_NO_INCREMENT);
3658 * If this is a periodic timer, re-arm it
3659 * so it will fire again. We do this before
3660 * calling any deferred procedure calls because
3661 * it's possible the DPC might cancel the timer,
3662 * in which case it would be wrong for us to
3663 * re-arm it again afterwards.
3666 if (timer->k_period) {
3668 tv.tv_usec = timer->k_period * 1000;
3669 timer->k_header.dh_inserted = TRUE;
3670 ntoskrnl_insert_timer(timer, tvtohz(&tv));
3671 #ifdef NTOSKRNL_DEBUG_TIMERS
3672 ntoskrnl_timer_reloads++;
3678 mtx_unlock(&ntoskrnl_dispatchlock);
3680 /* If there's a DPC associated with the timer, queue it up. */
3683 KeInsertQueueDpc(dpc, NULL, NULL);
3686 #ifdef NTOSKRNL_DEBUG_TIMERS
3688 sysctl_show_timers(SYSCTL_HANDLER_ARGS)
3693 ntoskrnl_show_timers();
3694 return (sysctl_handle_int(oidp, &ret, 0, req));
3698 ntoskrnl_show_timers()
3703 mtx_lock_spin(&ntoskrnl_calllock);
3704 l = ntoskrnl_calllist.nle_flink;
3705 while(l != &ntoskrnl_calllist) {
3709 mtx_unlock_spin(&ntoskrnl_calllock);
3712 printf("%d timers available (out of %d)\n", i, NTOSKRNL_TIMEOUTS);
3713 printf("timer sets: %qu\n", ntoskrnl_timer_sets);
3714 printf("timer reloads: %qu\n", ntoskrnl_timer_reloads);
3715 printf("timer cancels: %qu\n", ntoskrnl_timer_cancels);
3716 printf("timer fires: %qu\n", ntoskrnl_timer_fires);
3722 * Must be called with dispatcher lock held.
3726 ntoskrnl_insert_timer(timer, ticks)
3735 * Try and allocate a timer.
3737 mtx_lock_spin(&ntoskrnl_calllock);
3738 if (IsListEmpty(&ntoskrnl_calllist)) {
3739 mtx_unlock_spin(&ntoskrnl_calllock);
3740 #ifdef NTOSKRNL_DEBUG_TIMERS
3741 ntoskrnl_show_timers();
3743 panic("out of timers!");
3745 l = RemoveHeadList(&ntoskrnl_calllist);
3746 mtx_unlock_spin(&ntoskrnl_calllock);
3748 e = CONTAINING_RECORD(l, callout_entry, ce_list);
3751 timer->k_callout = c;
3753 callout_init(c, CALLOUT_MPSAFE);
3754 callout_reset(c, ticks, ntoskrnl_timercall, timer);
3758 ntoskrnl_remove_timer(timer)
3763 e = (callout_entry *)timer->k_callout;
3764 callout_stop(timer->k_callout);
3766 mtx_lock_spin(&ntoskrnl_calllock);
3767 InsertHeadList((&ntoskrnl_calllist), (&e->ce_list));
3768 mtx_unlock_spin(&ntoskrnl_calllock);
3772 KeInitializeTimer(timer)
3778 KeInitializeTimerEx(timer, EVENT_TYPE_NOTIFY);
3782 KeInitializeTimerEx(timer, type)
3789 bzero((char *)timer, sizeof(ktimer));
3790 InitializeListHead((&timer->k_header.dh_waitlisthead));
3791 timer->k_header.dh_sigstate = FALSE;
3792 timer->k_header.dh_inserted = FALSE;
3793 if (type == EVENT_TYPE_NOTIFY)
3794 timer->k_header.dh_type = DISP_TYPE_NOTIFICATION_TIMER;
3796 timer->k_header.dh_type = DISP_TYPE_SYNCHRONIZATION_TIMER;
3797 timer->k_header.dh_size = sizeof(ktimer) / sizeof(uint32_t);
3801 * DPC subsystem. A Windows Defered Procedure Call has the following
3803 * - It runs at DISPATCH_LEVEL.
3804 * - It can have one of 3 importance values that control when it
3805 * runs relative to other DPCs in the queue.
3806 * - On SMP systems, it can be set to run on a specific processor.
3807 * In order to satisfy the last property, we create a DPC thread for
3808 * each CPU in the system and bind it to that CPU. Each thread
3809 * maintains three queues with different importance levels, which
3810 * will be processed in order from lowest to highest.
3812 * In Windows, interrupt handlers run as DPCs. (Not to be confused
3813 * with ISRs, which run in interrupt context and can preempt DPCs.)
3814 * ISRs are given the highest importance so that they'll take
3815 * precedence over timers and other things.
3819 ntoskrnl_dpc_thread(arg)
3829 InitializeListHead(&kq->kq_disp);
3830 kq->kq_td = curthread;
3832 kq->kq_running = FALSE;
3833 KeInitializeSpinLock(&kq->kq_lock);
3834 KeInitializeEvent(&kq->kq_proc, EVENT_TYPE_SYNC, FALSE);
3835 KeInitializeEvent(&kq->kq_done, EVENT_TYPE_SYNC, FALSE);
3838 * Elevate our priority. DPCs are used to run interrupt
3839 * handlers, and they should trigger as soon as possible
3840 * once scheduled by an ISR.
3843 thread_lock(curthread);
3844 #ifdef NTOSKRNL_MULTIPLE_DPCS
3845 sched_bind(curthread, kq->kq_cpu);
3847 sched_prio(curthread, PRI_MIN_KERN);
3848 thread_unlock(curthread);
3851 KeWaitForSingleObject(&kq->kq_proc, 0, 0, TRUE, NULL);
3853 KeAcquireSpinLock(&kq->kq_lock, &irql);
3857 KeReleaseSpinLock(&kq->kq_lock, irql);
3861 kq->kq_running = TRUE;
3863 while (!IsListEmpty(&kq->kq_disp)) {
3864 l = RemoveHeadList((&kq->kq_disp));
3865 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3866 InitializeListHead((&d->k_dpclistentry));
3867 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
3868 MSCALL4(d->k_deferedfunc, d, d->k_deferredctx,
3869 d->k_sysarg1, d->k_sysarg2);
3870 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3873 kq->kq_running = FALSE;
3875 KeReleaseSpinLock(&kq->kq_lock, irql);
3877 KeSetEvent(&kq->kq_done, IO_NO_INCREMENT, FALSE);
3881 return; /* notreached */
3885 ntoskrnl_destroy_dpc_threads(void)
3892 #ifdef NTOSKRNL_MULTIPLE_DPCS
3893 for (i = 0; i < mp_ncpus; i++) {
3895 for (i = 0; i < 1; i++) {
3900 KeInitializeDpc(&dpc, NULL, NULL);
3901 KeSetTargetProcessorDpc(&dpc, i);
3902 KeInsertQueueDpc(&dpc, NULL, NULL);
3904 tsleep(kq->kq_td->td_proc, PWAIT, "dpcw", hz/10);
3909 ntoskrnl_insert_dpc(head, dpc)
3916 l = head->nle_flink;
3918 d = CONTAINING_RECORD(l, kdpc, k_dpclistentry);
3924 if (dpc->k_importance == KDPC_IMPORTANCE_LOW)
3925 InsertTailList((head), (&dpc->k_dpclistentry));
3927 InsertHeadList((head), (&dpc->k_dpclistentry));
3933 KeInitializeDpc(dpc, dpcfunc, dpcctx)
3942 dpc->k_deferedfunc = dpcfunc;
3943 dpc->k_deferredctx = dpcctx;
3944 dpc->k_num = KDPC_CPU_DEFAULT;
3945 dpc->k_importance = KDPC_IMPORTANCE_MEDIUM;
3946 InitializeListHead((&dpc->k_dpclistentry));
3950 KeInsertQueueDpc(dpc, sysarg1, sysarg2)
3964 #ifdef NTOSKRNL_MULTIPLE_DPCS
3965 KeRaiseIrql(DISPATCH_LEVEL, &irql);
3968 * By default, the DPC is queued to run on the same CPU
3969 * that scheduled it.
3972 if (dpc->k_num == KDPC_CPU_DEFAULT)
3973 kq += curthread->td_oncpu;
3976 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
3978 KeAcquireSpinLock(&kq->kq_lock, &irql);
3981 r = ntoskrnl_insert_dpc(&kq->kq_disp, dpc);
3983 dpc->k_sysarg1 = sysarg1;
3984 dpc->k_sysarg2 = sysarg2;
3986 KeReleaseSpinLock(&kq->kq_lock, irql);
3991 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
3997 KeRemoveQueueDpc(dpc)
4006 #ifdef NTOSKRNL_MULTIPLE_DPCS
4007 KeRaiseIrql(DISPATCH_LEVEL, &irql);
4009 kq = kq_queues + dpc->k_num;
4011 KeAcquireSpinLockAtDpcLevel(&kq->kq_lock);
4014 KeAcquireSpinLock(&kq->kq_lock, &irql);
4017 if (dpc->k_dpclistentry.nle_flink == &dpc->k_dpclistentry) {
4018 KeReleaseSpinLockFromDpcLevel(&kq->kq_lock);
4023 RemoveEntryList((&dpc->k_dpclistentry));
4024 InitializeListHead((&dpc->k_dpclistentry));
4026 KeReleaseSpinLock(&kq->kq_lock, irql);
4032 KeSetImportanceDpc(dpc, imp)
4036 if (imp != KDPC_IMPORTANCE_LOW &&
4037 imp != KDPC_IMPORTANCE_MEDIUM &&
4038 imp != KDPC_IMPORTANCE_HIGH)
4041 dpc->k_importance = (uint8_t)imp;
4045 KeSetTargetProcessorDpc(kdpc *dpc, uint8_t cpu)
4054 KeFlushQueuedDpcs(void)
4060 * Poke each DPC queue and wait
4061 * for them to drain.
4064 #ifdef NTOSKRNL_MULTIPLE_DPCS
4065 for (i = 0; i < mp_ncpus; i++) {
4067 for (i = 0; i < 1; i++) {
4070 KeSetEvent(&kq->kq_proc, IO_NO_INCREMENT, FALSE);
4071 KeWaitForSingleObject(&kq->kq_done, 0, 0, TRUE, NULL);
4076 KeGetCurrentProcessorNumber(void)
4078 return ((uint32_t)curthread->td_oncpu);
4082 KeSetTimerEx(timer, duetime, period, dpc)
4095 mtx_lock(&ntoskrnl_dispatchlock);
4097 if (timer->k_header.dh_inserted == TRUE) {
4098 ntoskrnl_remove_timer(timer);
4099 #ifdef NTOSKRNL_DEBUG_TIMERS
4100 ntoskrnl_timer_cancels++;
4102 timer->k_header.dh_inserted = FALSE;
4107 timer->k_duetime = duetime;
4108 timer->k_period = period;
4109 timer->k_header.dh_sigstate = FALSE;
4113 tv.tv_sec = - (duetime) / 10000000;
4114 tv.tv_usec = (- (duetime) / 10) -
4115 (tv.tv_sec * 1000000);
4117 ntoskrnl_time(&curtime);
4118 if (duetime < curtime)
4119 tv.tv_sec = tv.tv_usec = 0;
4121 tv.tv_sec = ((duetime) - curtime) / 10000000;
4122 tv.tv_usec = ((duetime) - curtime) / 10 -
4123 (tv.tv_sec * 1000000);
4127 timer->k_header.dh_inserted = TRUE;
4128 ntoskrnl_insert_timer(timer, tvtohz(&tv));
4129 #ifdef NTOSKRNL_DEBUG_TIMERS
4130 ntoskrnl_timer_sets++;
4133 mtx_unlock(&ntoskrnl_dispatchlock);
4139 KeSetTimer(timer, duetime, dpc)
4144 return (KeSetTimerEx(timer, duetime, 0, dpc));
4148 * The Windows DDK documentation seems to say that cancelling
4149 * a timer that has a DPC will result in the DPC also being
4150 * cancelled, but this isn't really the case.
4154 KeCancelTimer(timer)
4162 mtx_lock(&ntoskrnl_dispatchlock);
4164 pending = timer->k_header.dh_inserted;
4166 if (timer->k_header.dh_inserted == TRUE) {
4167 timer->k_header.dh_inserted = FALSE;
4168 ntoskrnl_remove_timer(timer);
4169 #ifdef NTOSKRNL_DEBUG_TIMERS
4170 ntoskrnl_timer_cancels++;
4174 mtx_unlock(&ntoskrnl_dispatchlock);
4180 KeReadStateTimer(timer)
4183 return (timer->k_header.dh_sigstate);
4187 KeDelayExecutionThread(uint8_t wait_mode, uint8_t alertable, int64_t *interval)
4192 panic("invalid wait_mode %d", wait_mode);
4194 KeInitializeTimer(&timer);
4195 KeSetTimer(&timer, *interval, NULL);
4196 KeWaitForSingleObject(&timer, 0, 0, alertable, NULL);
4198 return STATUS_SUCCESS;
4202 KeQueryInterruptTime(void)
4207 getmicrouptime(&tv);
4209 ticks = tvtohz(&tv);
4211 return ticks * ((10000000 + hz - 1) / hz);
4214 static struct thread *
4215 KeGetCurrentThread(void)
4222 KeSetPriorityThread(td, pri)
4229 return LOW_REALTIME_PRIORITY;
4231 if (td->td_priority <= PRI_MIN_KERN)
4232 old = HIGH_PRIORITY;
4233 else if (td->td_priority >= PRI_MAX_KERN)
4236 old = LOW_REALTIME_PRIORITY;
4239 if (pri == HIGH_PRIORITY)
4240 sched_prio(td, PRI_MIN_KERN);
4241 if (pri == LOW_REALTIME_PRIORITY)
4242 sched_prio(td, PRI_MIN_KERN + (PRI_MAX_KERN - PRI_MIN_KERN) / 2);
4243 if (pri == LOW_PRIORITY)
4244 sched_prio(td, PRI_MAX_KERN);
4253 printf("ntoskrnl dummy called...\n");
4257 image_patch_table ntoskrnl_functbl[] = {
4258 IMPORT_SFUNC(RtlZeroMemory, 2),
4259 IMPORT_SFUNC(RtlSecureZeroMemory, 2),
4260 IMPORT_SFUNC(RtlFillMemory, 3),
4261 IMPORT_SFUNC(RtlMoveMemory, 3),
4262 IMPORT_SFUNC(RtlCharToInteger, 3),
4263 IMPORT_SFUNC(RtlCopyMemory, 3),
4264 IMPORT_SFUNC(RtlCopyString, 2),
4265 IMPORT_SFUNC(RtlCompareMemory, 3),
4266 IMPORT_SFUNC(RtlEqualUnicodeString, 3),
4267 IMPORT_SFUNC(RtlCopyUnicodeString, 2),
4268 IMPORT_SFUNC(RtlUnicodeStringToAnsiString, 3),
4269 IMPORT_SFUNC(RtlAnsiStringToUnicodeString, 3),
4270 IMPORT_SFUNC(RtlInitAnsiString, 2),
4271 IMPORT_SFUNC_MAP(RtlInitString, RtlInitAnsiString, 2),
4272 IMPORT_SFUNC(RtlInitUnicodeString, 2),
4273 IMPORT_SFUNC(RtlFreeAnsiString, 1),
4274 IMPORT_SFUNC(RtlFreeUnicodeString, 1),
4275 IMPORT_SFUNC(RtlUnicodeStringToInteger, 3),
4276 IMPORT_CFUNC(sprintf, 0),
4277 IMPORT_CFUNC(vsprintf, 0),
4278 IMPORT_CFUNC_MAP(_snprintf, snprintf, 0),
4279 IMPORT_CFUNC_MAP(_vsnprintf, vsnprintf, 0),
4280 IMPORT_CFUNC(DbgPrint, 0),
4281 IMPORT_SFUNC(DbgBreakPoint, 0),
4282 IMPORT_SFUNC(KeBugCheckEx, 5),
4283 IMPORT_CFUNC(strncmp, 0),
4284 IMPORT_CFUNC(strcmp, 0),
4285 IMPORT_CFUNC_MAP(stricmp, strcasecmp, 0),
4286 IMPORT_CFUNC(strncpy, 0),
4287 IMPORT_CFUNC(strcpy, 0),
4288 IMPORT_CFUNC(strlen, 0),
4289 IMPORT_CFUNC_MAP(toupper, ntoskrnl_toupper, 0),
4290 IMPORT_CFUNC_MAP(tolower, ntoskrnl_tolower, 0),
4291 IMPORT_CFUNC_MAP(strstr, ntoskrnl_strstr, 0),
4292 IMPORT_CFUNC_MAP(strncat, ntoskrnl_strncat, 0),
4293 IMPORT_CFUNC_MAP(strchr, index, 0),
4294 IMPORT_CFUNC_MAP(strrchr, rindex, 0),
4295 IMPORT_CFUNC(memcpy, 0),
4296 IMPORT_CFUNC_MAP(memmove, ntoskrnl_memmove, 0),
4297 IMPORT_CFUNC_MAP(memset, ntoskrnl_memset, 0),
4298 IMPORT_CFUNC_MAP(memchr, ntoskrnl_memchr, 0),
4299 IMPORT_SFUNC(IoAllocateDriverObjectExtension, 4),
4300 IMPORT_SFUNC(IoGetDriverObjectExtension, 2),
4301 IMPORT_FFUNC(IofCallDriver, 2),
4302 IMPORT_FFUNC(IofCompleteRequest, 2),
4303 IMPORT_SFUNC(IoAcquireCancelSpinLock, 1),
4304 IMPORT_SFUNC(IoReleaseCancelSpinLock, 1),
4305 IMPORT_SFUNC(IoCancelIrp, 1),
4306 IMPORT_SFUNC(IoConnectInterrupt, 11),
4307 IMPORT_SFUNC(IoDisconnectInterrupt, 1),
4308 IMPORT_SFUNC(IoCreateDevice, 7),
4309 IMPORT_SFUNC(IoDeleteDevice, 1),
4310 IMPORT_SFUNC(IoGetAttachedDevice, 1),
4311 IMPORT_SFUNC(IoAttachDeviceToDeviceStack, 2),
4312 IMPORT_SFUNC(IoDetachDevice, 1),
4313 IMPORT_SFUNC(IoBuildSynchronousFsdRequest, 7),
4314 IMPORT_SFUNC(IoBuildAsynchronousFsdRequest, 6),
4315 IMPORT_SFUNC(IoBuildDeviceIoControlRequest, 9),
4316 IMPORT_SFUNC(IoAllocateIrp, 2),
4317 IMPORT_SFUNC(IoReuseIrp, 2),
4318 IMPORT_SFUNC(IoMakeAssociatedIrp, 2),
4319 IMPORT_SFUNC(IoFreeIrp, 1),
4320 IMPORT_SFUNC(IoInitializeIrp, 3),
4321 IMPORT_SFUNC(KeAcquireInterruptSpinLock, 1),
4322 IMPORT_SFUNC(KeReleaseInterruptSpinLock, 2),
4323 IMPORT_SFUNC(KeSynchronizeExecution, 3),
4324 IMPORT_SFUNC(KeWaitForSingleObject, 5),
4325 IMPORT_SFUNC(KeWaitForMultipleObjects, 8),
4326 IMPORT_SFUNC(_allmul, 4),
4327 IMPORT_SFUNC(_alldiv, 4),
4328 IMPORT_SFUNC(_allrem, 4),
4329 IMPORT_RFUNC(_allshr, 0),
4330 IMPORT_RFUNC(_allshl, 0),
4331 IMPORT_SFUNC(_aullmul, 4),
4332 IMPORT_SFUNC(_aulldiv, 4),
4333 IMPORT_SFUNC(_aullrem, 4),
4334 IMPORT_RFUNC(_aullshr, 0),
4335 IMPORT_RFUNC(_aullshl, 0),
4336 IMPORT_CFUNC(atoi, 0),
4337 IMPORT_CFUNC(atol, 0),
4338 IMPORT_CFUNC(rand, 0),
4339 IMPORT_CFUNC(srand, 0),
4340 IMPORT_SFUNC(WRITE_REGISTER_USHORT, 2),
4341 IMPORT_SFUNC(READ_REGISTER_USHORT, 1),
4342 IMPORT_SFUNC(WRITE_REGISTER_ULONG, 2),
4343 IMPORT_SFUNC(READ_REGISTER_ULONG, 1),
4344 IMPORT_SFUNC(READ_REGISTER_UCHAR, 1),
4345 IMPORT_SFUNC(WRITE_REGISTER_UCHAR, 2),
4346 IMPORT_SFUNC(ExInitializePagedLookasideList, 7),
4347 IMPORT_SFUNC(ExDeletePagedLookasideList, 1),
4348 IMPORT_SFUNC(ExInitializeNPagedLookasideList, 7),
4349 IMPORT_SFUNC(ExDeleteNPagedLookasideList, 1),
4350 IMPORT_FFUNC(InterlockedPopEntrySList, 1),
4351 IMPORT_FFUNC(InitializeSListHead, 1),
4352 IMPORT_FFUNC(InterlockedPushEntrySList, 2),
4353 IMPORT_SFUNC(ExQueryDepthSList, 1),
4354 IMPORT_FFUNC_MAP(ExpInterlockedPopEntrySList,
4355 InterlockedPopEntrySList, 1),
4356 IMPORT_FFUNC_MAP(ExpInterlockedPushEntrySList,
4357 InterlockedPushEntrySList, 2),
4358 IMPORT_FFUNC(ExInterlockedPopEntrySList, 2),
4359 IMPORT_FFUNC(ExInterlockedPushEntrySList, 3),
4360 IMPORT_SFUNC(ExAllocatePoolWithTag, 3),
4361 IMPORT_SFUNC(ExFreePoolWithTag, 2),
4362 IMPORT_SFUNC(ExFreePool, 1),
4364 IMPORT_FFUNC(KefAcquireSpinLockAtDpcLevel, 1),
4365 IMPORT_FFUNC(KefReleaseSpinLockFromDpcLevel,1),
4366 IMPORT_FFUNC(KeAcquireSpinLockRaiseToDpc, 1),
4369 * For AMD64, we can get away with just mapping
4370 * KeAcquireSpinLockRaiseToDpc() directly to KfAcquireSpinLock()
4371 * because the calling conventions end up being the same.
4372 * On i386, we have to be careful because KfAcquireSpinLock()
4373 * is _fastcall but KeAcquireSpinLockRaiseToDpc() isn't.
4375 IMPORT_SFUNC(KeAcquireSpinLockAtDpcLevel, 1),
4376 IMPORT_SFUNC(KeReleaseSpinLockFromDpcLevel, 1),
4377 IMPORT_SFUNC_MAP(KeAcquireSpinLockRaiseToDpc, KfAcquireSpinLock, 1),
4379 IMPORT_SFUNC_MAP(KeReleaseSpinLock, KfReleaseSpinLock, 1),
4380 IMPORT_FFUNC(InterlockedIncrement, 1),
4381 IMPORT_FFUNC(InterlockedDecrement, 1),
4382 IMPORT_FFUNC(InterlockedExchange, 2),
4383 IMPORT_FFUNC(ExInterlockedAddLargeStatistic, 2),
4384 IMPORT_SFUNC(IoAllocateMdl, 5),
4385 IMPORT_SFUNC(IoFreeMdl, 1),
4386 IMPORT_SFUNC(MmAllocateContiguousMemory, 2 + 1),
4387 IMPORT_SFUNC(MmAllocateContiguousMemorySpecifyCache, 5 + 3),
4388 IMPORT_SFUNC(MmFreeContiguousMemory, 1),
4389 IMPORT_SFUNC(MmFreeContiguousMemorySpecifyCache, 3),
4390 IMPORT_SFUNC(MmSizeOfMdl, 1),
4391 IMPORT_SFUNC(MmMapLockedPages, 2),
4392 IMPORT_SFUNC(MmMapLockedPagesSpecifyCache, 6),
4393 IMPORT_SFUNC(MmUnmapLockedPages, 2),
4394 IMPORT_SFUNC(MmBuildMdlForNonPagedPool, 1),
4395 IMPORT_SFUNC(MmGetPhysicalAddress, 1),
4396 IMPORT_SFUNC(MmGetSystemRoutineAddress, 1),
4397 IMPORT_SFUNC(MmIsAddressValid, 1),
4398 IMPORT_SFUNC(MmMapIoSpace, 3 + 1),
4399 IMPORT_SFUNC(MmUnmapIoSpace, 2),
4400 IMPORT_SFUNC(KeInitializeSpinLock, 1),
4401 IMPORT_SFUNC(IoIsWdmVersionAvailable, 2),
4402 IMPORT_SFUNC(IoOpenDeviceRegistryKey, 4),
4403 IMPORT_SFUNC(IoGetDeviceObjectPointer, 4),
4404 IMPORT_SFUNC(IoGetDeviceProperty, 5),
4405 IMPORT_SFUNC(IoAllocateWorkItem, 1),
4406 IMPORT_SFUNC(IoFreeWorkItem, 1),
4407 IMPORT_SFUNC(IoQueueWorkItem, 4),
4408 IMPORT_SFUNC(ExQueueWorkItem, 2),
4409 IMPORT_SFUNC(ntoskrnl_workitem, 2),
4410 IMPORT_SFUNC(KeInitializeMutex, 2),
4411 IMPORT_SFUNC(KeReleaseMutex, 2),
4412 IMPORT_SFUNC(KeReadStateMutex, 1),
4413 IMPORT_SFUNC(KeInitializeEvent, 3),
4414 IMPORT_SFUNC(KeSetEvent, 3),
4415 IMPORT_SFUNC(KeResetEvent, 1),
4416 IMPORT_SFUNC(KeClearEvent, 1),
4417 IMPORT_SFUNC(KeReadStateEvent, 1),
4418 IMPORT_SFUNC(KeInitializeTimer, 1),
4419 IMPORT_SFUNC(KeInitializeTimerEx, 2),
4420 IMPORT_SFUNC(KeSetTimer, 3),
4421 IMPORT_SFUNC(KeSetTimerEx, 4),
4422 IMPORT_SFUNC(KeCancelTimer, 1),
4423 IMPORT_SFUNC(KeReadStateTimer, 1),
4424 IMPORT_SFUNC(KeInitializeDpc, 3),
4425 IMPORT_SFUNC(KeInsertQueueDpc, 3),
4426 IMPORT_SFUNC(KeRemoveQueueDpc, 1),
4427 IMPORT_SFUNC(KeSetImportanceDpc, 2),
4428 IMPORT_SFUNC(KeSetTargetProcessorDpc, 2),
4429 IMPORT_SFUNC(KeFlushQueuedDpcs, 0),
4430 IMPORT_SFUNC(KeGetCurrentProcessorNumber, 1),
4431 IMPORT_SFUNC(ObReferenceObjectByHandle, 6),
4432 IMPORT_FFUNC(ObfDereferenceObject, 1),
4433 IMPORT_SFUNC(ZwClose, 1),
4434 IMPORT_SFUNC(PsCreateSystemThread, 7),
4435 IMPORT_SFUNC(PsTerminateSystemThread, 1),
4436 IMPORT_SFUNC(IoWMIRegistrationControl, 2),
4437 IMPORT_SFUNC(WmiQueryTraceInformation, 5),
4438 IMPORT_CFUNC(WmiTraceMessage, 0),
4439 IMPORT_SFUNC(KeQuerySystemTime, 1),
4440 IMPORT_CFUNC(KeTickCount, 0),
4441 IMPORT_SFUNC(KeDelayExecutionThread, 3),
4442 IMPORT_SFUNC(KeQueryInterruptTime, 0),
4443 IMPORT_SFUNC(KeGetCurrentThread, 0),
4444 IMPORT_SFUNC(KeSetPriorityThread, 2),
4447 * This last entry is a catch-all for any function we haven't
4448 * implemented yet. The PE import list patching routine will
4449 * use it for any function that doesn't have an explicit match
4453 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
4457 { NULL, NULL, NULL }