1 /*-
2 * Copyright (c) 2003
3 * Bill Paul <wpaul@windriver.com>. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
19 *
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.
31 */
32
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD: releng/6.0/sys/compat/ndis/subr_hal.c 151742 2005-10-27 17:08:57Z wpaul $");
35
36 #include <sys/param.h>
37 #include <sys/types.h>
38 #include <sys/errno.h>
39
40 #include <sys/callout.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/sched.h>
46 #include <sys/module.h>
47
48 #include <sys/systm.h>
49 #include <machine/clock.h>
50 #include <machine/bus.h>
51
52 #include <sys/bus.h>
53 #include <sys/rman.h>
54
55 #include <compat/ndis/pe_var.h>
56 #include <compat/ndis/resource_var.h>
57 #include <compat/ndis/cfg_var.h>
58 #include <compat/ndis/ntoskrnl_var.h>
59 #include <compat/ndis/hal_var.h>
60
61 static void KeStallExecutionProcessor(uint32_t);
62 static void WRITE_PORT_BUFFER_ULONG(uint32_t *,
63 uint32_t *, uint32_t);
64 static void WRITE_PORT_BUFFER_USHORT(uint16_t *,
65 uint16_t *, uint32_t);
66 static void WRITE_PORT_BUFFER_UCHAR(uint8_t *,
67 uint8_t *, uint32_t);
68 static void WRITE_PORT_ULONG(uint32_t *, uint32_t);
69 static void WRITE_PORT_USHORT(uint16_t *, uint16_t);
70 static void WRITE_PORT_UCHAR(uint8_t *, uint8_t);
71 static uint32_t READ_PORT_ULONG(uint32_t *);
72 static uint16_t READ_PORT_USHORT(uint16_t *);
73 static uint8_t READ_PORT_UCHAR(uint8_t *);
74 static void READ_PORT_BUFFER_ULONG(uint32_t *,
75 uint32_t *, uint32_t);
76 static void READ_PORT_BUFFER_USHORT(uint16_t *,
77 uint16_t *, uint32_t);
78 static void READ_PORT_BUFFER_UCHAR(uint8_t *,
79 uint8_t *, uint32_t);
80 static uint64_t KeQueryPerformanceCounter(uint64_t *);
81 static void dummy (void);
82
83 #define NDIS_MAXCPUS 64
84 static struct mtx disp_lock[NDIS_MAXCPUS];
85
86 int
87 hal_libinit()
88 {
89 image_patch_table *patch;
90 int i;
91
92 for (i = 0; i < NDIS_MAXCPUS; i++)
93 mtx_init(&disp_lock[i], "HAL preemption lock",
94 "HAL lock", MTX_RECURSE|MTX_DEF);
95
96 patch = hal_functbl;
97 while (patch->ipt_func != NULL) {
98 windrv_wrap((funcptr)patch->ipt_func,
99 (funcptr *)&patch->ipt_wrap,
100 patch->ipt_argcnt, patch->ipt_ftype);
101 patch++;
102 }
103
104
105 return(0);
106 }
107
108 int
109 hal_libfini()
110 {
111 image_patch_table *patch;
112 int i;
113
114 for (i = 0; i < NDIS_MAXCPUS; i++)
115 mtx_destroy(&disp_lock[i]);
116
117 patch = hal_functbl;
118 while (patch->ipt_func != NULL) {
119 windrv_unwrap(patch->ipt_wrap);
120 patch++;
121 }
122
123 return(0);
124 }
125
126 static void
127 KeStallExecutionProcessor(usecs)
128 uint32_t usecs;
129 {
130 DELAY(usecs);
131 return;
132 }
133
134 static void
135 WRITE_PORT_ULONG(port, val)
136 uint32_t *port;
137 uint32_t val;
138 {
139 bus_space_write_4(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
140 return;
141 }
142
143 static void
144 WRITE_PORT_USHORT(port, val)
145 uint16_t *port;
146 uint16_t val;
147 {
148 bus_space_write_2(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
149 return;
150 }
151
152 static void
153 WRITE_PORT_UCHAR(port, val)
154 uint8_t *port;
155 uint8_t val;
156 {
157 bus_space_write_1(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
158 return;
159 }
160
161 static void
162 WRITE_PORT_BUFFER_ULONG(port, val, cnt)
163 uint32_t *port;
164 uint32_t *val;
165 uint32_t cnt;
166 {
167 bus_space_write_multi_4(NDIS_BUS_SPACE_IO, 0x0,
168 (bus_size_t)port, val, cnt);
169 return;
170 }
171
172 static void
173 WRITE_PORT_BUFFER_USHORT(port, val, cnt)
174 uint16_t *port;
175 uint16_t *val;
176 uint32_t cnt;
177 {
178 bus_space_write_multi_2(NDIS_BUS_SPACE_IO, 0x0,
179 (bus_size_t)port, val, cnt);
180 return;
181 }
182
183 static void
184 WRITE_PORT_BUFFER_UCHAR(port, val, cnt)
185 uint8_t *port;
186 uint8_t *val;
187 uint32_t cnt;
188 {
189 bus_space_write_multi_1(NDIS_BUS_SPACE_IO, 0x0,
190 (bus_size_t)port, val, cnt);
191 return;
192 }
193
194 static uint16_t
195 READ_PORT_USHORT(port)
196 uint16_t *port;
197 {
198 return(bus_space_read_2(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
199 }
200
201 static uint32_t
202 READ_PORT_ULONG(port)
203 uint32_t *port;
204 {
205 return(bus_space_read_4(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
206 }
207
208 static uint8_t
209 READ_PORT_UCHAR(port)
210 uint8_t *port;
211 {
212 return(bus_space_read_1(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
213 }
214
215 static void
216 READ_PORT_BUFFER_ULONG(port, val, cnt)
217 uint32_t *port;
218 uint32_t *val;
219 uint32_t cnt;
220 {
221 bus_space_read_multi_4(NDIS_BUS_SPACE_IO, 0x0,
222 (bus_size_t)port, val, cnt);
223 return;
224 }
225
226 static void
227 READ_PORT_BUFFER_USHORT(port, val, cnt)
228 uint16_t *port;
229 uint16_t *val;
230 uint32_t cnt;
231 {
232 bus_space_read_multi_2(NDIS_BUS_SPACE_IO, 0x0,
233 (bus_size_t)port, val, cnt);
234 return;
235 }
236
237 static void
238 READ_PORT_BUFFER_UCHAR(port, val, cnt)
239 uint8_t *port;
240 uint8_t *val;
241 uint32_t cnt;
242 {
243 bus_space_read_multi_1(NDIS_BUS_SPACE_IO, 0x0,
244 (bus_size_t)port, val, cnt);
245 return;
246 }
247
248 /*
249 * The spinlock implementation in Windows differs from that of FreeBSD.
250 * The basic operation of spinlocks involves two steps: 1) spin in a
251 * tight loop while trying to acquire a lock, 2) after obtaining the
252 * lock, disable preemption. (Note that on uniprocessor systems, you're
253 * allowed to skip the first step and just lock out pre-emption, since
254 * it's not possible for you to be in contention with another running
255 * thread.) Later, you release the lock then re-enable preemption.
256 * The difference between Windows and FreeBSD lies in how preemption
257 * is disabled. In FreeBSD, it's done using critical_enter(), which on
258 * the x86 arch translates to a cli instruction. This masks off all
259 * interrupts, and effectively stops the scheduler from ever running
260 * so _nothing_ can execute except the current thread. In Windows,
261 * preemption is disabled by raising the processor IRQL to DISPATCH_LEVEL.
262 * This stops other threads from running, but does _not_ block device
263 * interrupts. This means ISRs can still run, and they can make other
264 * threads runable, but those other threads won't be able to execute
265 * until the current thread lowers the IRQL to something less than
266 * DISPATCH_LEVEL.
267 *
268 * There's another commonly used IRQL in Windows, which is APC_LEVEL.
269 * An APC is an Asynchronous Procedure Call, which differs from a DPC
270 * (Defered Procedure Call) in that a DPC is queued up to run in
271 * another thread, while an APC runs in the thread that scheduled
272 * it (similar to a signal handler in a UNIX process). We don't
273 * actually support the notion of APCs in FreeBSD, so for now, the
274 * only IRQLs we're interested in are DISPATCH_LEVEL and PASSIVE_LEVEL.
275 *
276 * To simulate DISPATCH_LEVEL, we raise the current thread's priority
277 * to PI_REALTIME, which is the highest we can give it. This should,
278 * if I understand things correctly, prevent anything except for an
279 * interrupt thread from preempting us. PASSIVE_LEVEL is basically
280 * everything else.
281 *
282 * Be aware that, at least on the x86 arch, the Windows spinlock
283 * functions are divided up in peculiar ways. The actual spinlock
284 * functions are KfAcquireSpinLock() and KfReleaseSpinLock(), and
285 * they live in HAL.dll. Meanwhile, KeInitializeSpinLock(),
286 * KefAcquireSpinLockAtDpcLevel() and KefReleaseSpinLockFromDpcLevel()
287 * live in ntoskrnl.exe. Most Windows source code will call
288 * KeAcquireSpinLock() and KeReleaseSpinLock(), but these are just
289 * macros that call KfAcquireSpinLock() and KfReleaseSpinLock().
290 * KefAcquireSpinLockAtDpcLevel() and KefReleaseSpinLockFromDpcLevel()
291 * perform the lock aquisition/release functions without doing the
292 * IRQL manipulation, and are used when one is already running at
293 * DISPATCH_LEVEL. Make sense? Good.
294 *
295 * According to the Microsoft documentation, any thread that calls
296 * KeAcquireSpinLock() must be running at IRQL <= DISPATCH_LEVEL. If
297 * we detect someone trying to acquire a spinlock from DEVICE_LEVEL
298 * or HIGH_LEVEL, we panic.
299 *
300 * Alternate sleep-lock-based spinlock implementation
301 * --------------------------------------------------
302 *
303 * The earlier spinlock implementation was arguably a bit of a hack
304 * and presented several problems. It was basically designed to provide
305 * the functionality of spinlocks without incurring the wrath of
306 * WITNESS. We could get away with using both our spinlock implementation
307 * and FreeBSD sleep locks at the same time, but if WITNESS knew what
308 * we were really up to, it would have spanked us rather severely.
309 *
310 * There's another method we can use based entirely on sleep locks.
311 * First, it's important to realize that everything we're locking
312 * resides inside Project Evil itself: any critical data being locked
313 * by drivers belongs to the drivers, and should not be referenced
314 * by any other OS code outside of the NDISulator. The priority-based
315 * locking scheme has system-wide effects, just like real spinlocks
316 * (blocking preemption affects the whole CPU), but since we keep all
317 * our critical data private, we can use a simpler mechanism that
318 * affects only code/threads directly related to Project Evil.
319 *
320 * The idea is to create a sleep lock mutex for each CPU in the system.
321 * When a CPU running in the NDISulator wants to acquire a spinlock, it
322 * does the following:
323 * - Pin ourselves to the current CPU
324 * - Acquire the mutex for the current CPU
325 * - Spin on the spinlock variable using atomic test and set, just like
326 * a real spinlock.
327 * - Once we have the lock, we execute our critical code
328 *
329 * To give up the lock, we do:
330 * - Clear the spinlock variable with an atomic op
331 * - Release the per-CPU mutex
332 * - Unpin ourselves from the current CPU.
333 *
334 * On a uniprocessor system, this means all threads that access protected
335 * data are serialized through the per-CPU mutex. After one thread
336 * acquires the 'spinlock,' any other thread that uses a spinlock on the
337 * current CPU will block on the per-CPU mutex, which has the same general
338 * effect of blocking pre-emption, but _only_ for those threads that are
339 * running NDISulator code.
340 *
341 * On a multiprocessor system, threads on different CPUs all block on
342 * their respective per-CPU mutex, and the atomic test/set operation
343 * on the spinlock variable provides inter-CPU synchronization, though
344 * only for threads running NDISulator code.
345 *
346 * This method solves an important problem. In Windows, you're allowed
347 * to do an ExAllocatePoolWithTag() with a spinlock held, provided you
348 * allocate from NonPagedPool. This implies an atomic heap allocation
349 * that will not cause the current thread to sleep. (You can't sleep
350 * while holding real spinlock: clowns will eat you.) But in FreeBSD,
351 * malloc(9) _always_ triggers the acquisition of a sleep lock, even
352 * when you use M_NOWAIT. This is not a problem for FreeBSD native
353 * code: you're allowed to sleep in things like interrupt threads. But
354 * it is a problem with the old priority-based spinlock implementation:
355 * even though we get away with it most of the time, we really can't
356 * do a malloc(9) after doing a KeAcquireSpinLock() or KeRaiseIrql().
357 * With the new implementation, it's not a problem: you're allowed to
358 * acquire more than one sleep lock (as long as you avoid lock order
359 * reversals).
360 *
361 * The one drawback to this approach is that now we have a lot of
362 * contention on one per-CPU mutex within the NDISulator code. Whether
363 * or not this is preferable to the expected Windows spinlock behavior
364 * of blocking pre-emption is debatable.
365 */
366
367 uint8_t
368 KfAcquireSpinLock(lock)
369 kspin_lock *lock;
370 {
371 uint8_t oldirql;
372
373 /* I am so going to hell for this. */
374 if (KeGetCurrentIrql() > DISPATCH_LEVEL)
375 panic("IRQL_NOT_LESS_THAN_OR_EQUAL");
376
377 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
378 KeAcquireSpinLockAtDpcLevel(lock);
379
380 return(oldirql);
381 }
382
383 void
384 KfReleaseSpinLock(lock, newirql)
385 kspin_lock *lock;
386 uint8_t newirql;
387 {
388 KeReleaseSpinLockFromDpcLevel(lock);
389 KeLowerIrql(newirql);
390
391 return;
392 }
393
394 uint8_t
395 KeGetCurrentIrql()
396 {
397 if (mtx_owned(&disp_lock[curthread->td_oncpu]))
398 return(DISPATCH_LEVEL);
399 return(PASSIVE_LEVEL);
400 }
401
402 static uint64_t
403 KeQueryPerformanceCounter(freq)
404 uint64_t *freq;
405 {
406 if (freq != NULL)
407 *freq = hz;
408
409 return((uint64_t)ticks);
410 }
411
412 uint8_t
413 KfRaiseIrql(irql)
414 uint8_t irql;
415 {
416 uint8_t oldirql;
417
418 oldirql = KeGetCurrentIrql();
419 if (irql < oldirql)
420 panic("IRQL_NOT_LESS_THAN");
421
422 if (oldirql != DISPATCH_LEVEL) {
423 sched_pin();
424 mtx_lock(&disp_lock[curthread->td_oncpu]);
425 }
426
427 return(oldirql);
428 }
429
430 void
431 KfLowerIrql(oldirql)
432 uint8_t oldirql;
433 {
434 if (oldirql == DISPATCH_LEVEL)
435 return;
436
437 if (KeGetCurrentIrql() != DISPATCH_LEVEL)
438 panic("IRQL_NOT_GREATER_THAN");
439
440 mtx_unlock(&disp_lock[curthread->td_oncpu]);
441 sched_unpin();
442
443 return;
444 }
445
446 static void dummy()
447 {
448 printf ("hal dummy called...\n");
449 return;
450 }
451
452 image_patch_table hal_functbl[] = {
453 IMPORT_SFUNC(KeStallExecutionProcessor, 1),
454 IMPORT_SFUNC(WRITE_PORT_ULONG, 2),
455 IMPORT_SFUNC(WRITE_PORT_USHORT, 2),
456 IMPORT_SFUNC(WRITE_PORT_UCHAR, 2),
457 IMPORT_SFUNC(WRITE_PORT_BUFFER_ULONG, 3),
458 IMPORT_SFUNC(WRITE_PORT_BUFFER_USHORT, 3),
459 IMPORT_SFUNC(WRITE_PORT_BUFFER_UCHAR, 3),
460 IMPORT_SFUNC(READ_PORT_ULONG, 1),
461 IMPORT_SFUNC(READ_PORT_USHORT, 1),
462 IMPORT_SFUNC(READ_PORT_UCHAR, 1),
463 IMPORT_SFUNC(READ_PORT_BUFFER_ULONG, 3),
464 IMPORT_SFUNC(READ_PORT_BUFFER_USHORT, 3),
465 IMPORT_SFUNC(READ_PORT_BUFFER_UCHAR, 3),
466 IMPORT_FFUNC(KfAcquireSpinLock, 1),
467 IMPORT_FFUNC(KfReleaseSpinLock, 1),
468 IMPORT_SFUNC(KeGetCurrentIrql, 0),
469 IMPORT_SFUNC(KeQueryPerformanceCounter, 1),
470 IMPORT_FFUNC(KfLowerIrql, 1),
471 IMPORT_FFUNC(KfRaiseIrql, 1),
472
473 /*
474 * This last entry is a catch-all for any function we haven't
475 * implemented yet. The PE import list patching routine will
476 * use it for any function that doesn't have an explicit match
477 * in this table.
478 */
479
480 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
481
482 /* End of list. */
483
484 { NULL, NULL, NULL }
485 };
Cache object: 269089d5f04e4640ec40a486d444ccc5
|