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: src/sys/compat/ndis/subr_hal.c,v 1.28 2006/05/16 14:37:57 phk Exp $");
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/bus.h>
50
51 #include <sys/bus.h>
52 #include <sys/rman.h>
53
54 #include <compat/ndis/pe_var.h>
55 #include <compat/ndis/resource_var.h>
56 #include <compat/ndis/cfg_var.h>
57 #include <compat/ndis/ntoskrnl_var.h>
58 #include <compat/ndis/hal_var.h>
59
60 static void KeStallExecutionProcessor(uint32_t);
61 static void WRITE_PORT_BUFFER_ULONG(uint32_t *,
62 uint32_t *, uint32_t);
63 static void WRITE_PORT_BUFFER_USHORT(uint16_t *,
64 uint16_t *, uint32_t);
65 static void WRITE_PORT_BUFFER_UCHAR(uint8_t *,
66 uint8_t *, uint32_t);
67 static void WRITE_PORT_ULONG(uint32_t *, uint32_t);
68 static void WRITE_PORT_USHORT(uint16_t *, uint16_t);
69 static void WRITE_PORT_UCHAR(uint8_t *, uint8_t);
70 static uint32_t READ_PORT_ULONG(uint32_t *);
71 static uint16_t READ_PORT_USHORT(uint16_t *);
72 static uint8_t READ_PORT_UCHAR(uint8_t *);
73 static void READ_PORT_BUFFER_ULONG(uint32_t *,
74 uint32_t *, uint32_t);
75 static void READ_PORT_BUFFER_USHORT(uint16_t *,
76 uint16_t *, uint32_t);
77 static void READ_PORT_BUFFER_UCHAR(uint8_t *,
78 uint8_t *, uint32_t);
79 static uint64_t KeQueryPerformanceCounter(uint64_t *);
80 static void _KeLowerIrql(uint8_t);
81 static uint8_t KeRaiseIrqlToDpcLevel(void);
82 static void dummy (void);
83
84 #define NDIS_MAXCPUS 64
85 static struct mtx disp_lock[NDIS_MAXCPUS];
86
87 int
88 hal_libinit()
89 {
90 image_patch_table *patch;
91 int i;
92
93 for (i = 0; i < NDIS_MAXCPUS; i++)
94 mtx_init(&disp_lock[i], "HAL preemption lock",
95 "HAL lock", MTX_RECURSE|MTX_DEF);
96
97 patch = hal_functbl;
98 while (patch->ipt_func != NULL) {
99 windrv_wrap((funcptr)patch->ipt_func,
100 (funcptr *)&patch->ipt_wrap,
101 patch->ipt_argcnt, patch->ipt_ftype);
102 patch++;
103 }
104
105
106 return(0);
107 }
108
109 int
110 hal_libfini()
111 {
112 image_patch_table *patch;
113 int i;
114
115 for (i = 0; i < NDIS_MAXCPUS; i++)
116 mtx_destroy(&disp_lock[i]);
117
118 patch = hal_functbl;
119 while (patch->ipt_func != NULL) {
120 windrv_unwrap(patch->ipt_wrap);
121 patch++;
122 }
123
124 return(0);
125 }
126
127 static void
128 KeStallExecutionProcessor(usecs)
129 uint32_t usecs;
130 {
131 DELAY(usecs);
132 return;
133 }
134
135 static void
136 WRITE_PORT_ULONG(port, val)
137 uint32_t *port;
138 uint32_t val;
139 {
140 bus_space_write_4(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
141 return;
142 }
143
144 static void
145 WRITE_PORT_USHORT(port, val)
146 uint16_t *port;
147 uint16_t val;
148 {
149 bus_space_write_2(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
150 return;
151 }
152
153 static void
154 WRITE_PORT_UCHAR(port, val)
155 uint8_t *port;
156 uint8_t val;
157 {
158 bus_space_write_1(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port, val);
159 return;
160 }
161
162 static void
163 WRITE_PORT_BUFFER_ULONG(port, val, cnt)
164 uint32_t *port;
165 uint32_t *val;
166 uint32_t cnt;
167 {
168 bus_space_write_multi_4(NDIS_BUS_SPACE_IO, 0x0,
169 (bus_size_t)port, val, cnt);
170 return;
171 }
172
173 static void
174 WRITE_PORT_BUFFER_USHORT(port, val, cnt)
175 uint16_t *port;
176 uint16_t *val;
177 uint32_t cnt;
178 {
179 bus_space_write_multi_2(NDIS_BUS_SPACE_IO, 0x0,
180 (bus_size_t)port, val, cnt);
181 return;
182 }
183
184 static void
185 WRITE_PORT_BUFFER_UCHAR(port, val, cnt)
186 uint8_t *port;
187 uint8_t *val;
188 uint32_t cnt;
189 {
190 bus_space_write_multi_1(NDIS_BUS_SPACE_IO, 0x0,
191 (bus_size_t)port, val, cnt);
192 return;
193 }
194
195 static uint16_t
196 READ_PORT_USHORT(port)
197 uint16_t *port;
198 {
199 return(bus_space_read_2(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
200 }
201
202 static uint32_t
203 READ_PORT_ULONG(port)
204 uint32_t *port;
205 {
206 return(bus_space_read_4(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
207 }
208
209 static uint8_t
210 READ_PORT_UCHAR(port)
211 uint8_t *port;
212 {
213 return(bus_space_read_1(NDIS_BUS_SPACE_IO, 0x0, (bus_size_t)port));
214 }
215
216 static void
217 READ_PORT_BUFFER_ULONG(port, val, cnt)
218 uint32_t *port;
219 uint32_t *val;
220 uint32_t cnt;
221 {
222 bus_space_read_multi_4(NDIS_BUS_SPACE_IO, 0x0,
223 (bus_size_t)port, val, cnt);
224 return;
225 }
226
227 static void
228 READ_PORT_BUFFER_USHORT(port, val, cnt)
229 uint16_t *port;
230 uint16_t *val;
231 uint32_t cnt;
232 {
233 bus_space_read_multi_2(NDIS_BUS_SPACE_IO, 0x0,
234 (bus_size_t)port, val, cnt);
235 return;
236 }
237
238 static void
239 READ_PORT_BUFFER_UCHAR(port, val, cnt)
240 uint8_t *port;
241 uint8_t *val;
242 uint32_t cnt;
243 {
244 bus_space_read_multi_1(NDIS_BUS_SPACE_IO, 0x0,
245 (bus_size_t)port, val, cnt);
246 return;
247 }
248
249 /*
250 * The spinlock implementation in Windows differs from that of FreeBSD.
251 * The basic operation of spinlocks involves two steps: 1) spin in a
252 * tight loop while trying to acquire a lock, 2) after obtaining the
253 * lock, disable preemption. (Note that on uniprocessor systems, you're
254 * allowed to skip the first step and just lock out pre-emption, since
255 * it's not possible for you to be in contention with another running
256 * thread.) Later, you release the lock then re-enable preemption.
257 * The difference between Windows and FreeBSD lies in how preemption
258 * is disabled. In FreeBSD, it's done using critical_enter(), which on
259 * the x86 arch translates to a cli instruction. This masks off all
260 * interrupts, and effectively stops the scheduler from ever running
261 * so _nothing_ can execute except the current thread. In Windows,
262 * preemption is disabled by raising the processor IRQL to DISPATCH_LEVEL.
263 * This stops other threads from running, but does _not_ block device
264 * interrupts. This means ISRs can still run, and they can make other
265 * threads runable, but those other threads won't be able to execute
266 * until the current thread lowers the IRQL to something less than
267 * DISPATCH_LEVEL.
268 *
269 * There's another commonly used IRQL in Windows, which is APC_LEVEL.
270 * An APC is an Asynchronous Procedure Call, which differs from a DPC
271 * (Defered Procedure Call) in that a DPC is queued up to run in
272 * another thread, while an APC runs in the thread that scheduled
273 * it (similar to a signal handler in a UNIX process). We don't
274 * actually support the notion of APCs in FreeBSD, so for now, the
275 * only IRQLs we're interested in are DISPATCH_LEVEL and PASSIVE_LEVEL.
276 *
277 * To simulate DISPATCH_LEVEL, we raise the current thread's priority
278 * to PI_REALTIME, which is the highest we can give it. This should,
279 * if I understand things correctly, prevent anything except for an
280 * interrupt thread from preempting us. PASSIVE_LEVEL is basically
281 * everything else.
282 *
283 * Be aware that, at least on the x86 arch, the Windows spinlock
284 * functions are divided up in peculiar ways. The actual spinlock
285 * functions are KfAcquireSpinLock() and KfReleaseSpinLock(), and
286 * they live in HAL.dll. Meanwhile, KeInitializeSpinLock(),
287 * KefAcquireSpinLockAtDpcLevel() and KefReleaseSpinLockFromDpcLevel()
288 * live in ntoskrnl.exe. Most Windows source code will call
289 * KeAcquireSpinLock() and KeReleaseSpinLock(), but these are just
290 * macros that call KfAcquireSpinLock() and KfReleaseSpinLock().
291 * KefAcquireSpinLockAtDpcLevel() and KefReleaseSpinLockFromDpcLevel()
292 * perform the lock aquisition/release functions without doing the
293 * IRQL manipulation, and are used when one is already running at
294 * DISPATCH_LEVEL. Make sense? Good.
295 *
296 * According to the Microsoft documentation, any thread that calls
297 * KeAcquireSpinLock() must be running at IRQL <= DISPATCH_LEVEL. If
298 * we detect someone trying to acquire a spinlock from DEVICE_LEVEL
299 * or HIGH_LEVEL, we panic.
300 *
301 * Alternate sleep-lock-based spinlock implementation
302 * --------------------------------------------------
303 *
304 * The earlier spinlock implementation was arguably a bit of a hack
305 * and presented several problems. It was basically designed to provide
306 * the functionality of spinlocks without incurring the wrath of
307 * WITNESS. We could get away with using both our spinlock implementation
308 * and FreeBSD sleep locks at the same time, but if WITNESS knew what
309 * we were really up to, it would have spanked us rather severely.
310 *
311 * There's another method we can use based entirely on sleep locks.
312 * First, it's important to realize that everything we're locking
313 * resides inside Project Evil itself: any critical data being locked
314 * by drivers belongs to the drivers, and should not be referenced
315 * by any other OS code outside of the NDISulator. The priority-based
316 * locking scheme has system-wide effects, just like real spinlocks
317 * (blocking preemption affects the whole CPU), but since we keep all
318 * our critical data private, we can use a simpler mechanism that
319 * affects only code/threads directly related to Project Evil.
320 *
321 * The idea is to create a sleep lock mutex for each CPU in the system.
322 * When a CPU running in the NDISulator wants to acquire a spinlock, it
323 * does the following:
324 * - Pin ourselves to the current CPU
325 * - Acquire the mutex for the current CPU
326 * - Spin on the spinlock variable using atomic test and set, just like
327 * a real spinlock.
328 * - Once we have the lock, we execute our critical code
329 *
330 * To give up the lock, we do:
331 * - Clear the spinlock variable with an atomic op
332 * - Release the per-CPU mutex
333 * - Unpin ourselves from the current CPU.
334 *
335 * On a uniprocessor system, this means all threads that access protected
336 * data are serialized through the per-CPU mutex. After one thread
337 * acquires the 'spinlock,' any other thread that uses a spinlock on the
338 * current CPU will block on the per-CPU mutex, which has the same general
339 * effect of blocking pre-emption, but _only_ for those threads that are
340 * running NDISulator code.
341 *
342 * On a multiprocessor system, threads on different CPUs all block on
343 * their respective per-CPU mutex, and the atomic test/set operation
344 * on the spinlock variable provides inter-CPU synchronization, though
345 * only for threads running NDISulator code.
346 *
347 * This method solves an important problem. In Windows, you're allowed
348 * to do an ExAllocatePoolWithTag() with a spinlock held, provided you
349 * allocate from NonPagedPool. This implies an atomic heap allocation
350 * that will not cause the current thread to sleep. (You can't sleep
351 * while holding real spinlock: clowns will eat you.) But in FreeBSD,
352 * malloc(9) _always_ triggers the acquisition of a sleep lock, even
353 * when you use M_NOWAIT. This is not a problem for FreeBSD native
354 * code: you're allowed to sleep in things like interrupt threads. But
355 * it is a problem with the old priority-based spinlock implementation:
356 * even though we get away with it most of the time, we really can't
357 * do a malloc(9) after doing a KeAcquireSpinLock() or KeRaiseIrql().
358 * With the new implementation, it's not a problem: you're allowed to
359 * acquire more than one sleep lock (as long as you avoid lock order
360 * reversals).
361 *
362 * The one drawback to this approach is that now we have a lot of
363 * contention on one per-CPU mutex within the NDISulator code. Whether
364 * or not this is preferable to the expected Windows spinlock behavior
365 * of blocking pre-emption is debatable.
366 */
367
368 uint8_t
369 KfAcquireSpinLock(lock)
370 kspin_lock *lock;
371 {
372 uint8_t oldirql;
373
374 KeRaiseIrql(DISPATCH_LEVEL, &oldirql);
375 KeAcquireSpinLockAtDpcLevel(lock);
376
377 return(oldirql);
378 }
379
380 void
381 KfReleaseSpinLock(lock, newirql)
382 kspin_lock *lock;
383 uint8_t newirql;
384 {
385 KeReleaseSpinLockFromDpcLevel(lock);
386 KeLowerIrql(newirql);
387
388 return;
389 }
390
391 uint8_t
392 KeGetCurrentIrql()
393 {
394 if (mtx_owned(&disp_lock[curthread->td_oncpu]))
395 return(DISPATCH_LEVEL);
396 return(PASSIVE_LEVEL);
397 }
398
399 static uint64_t
400 KeQueryPerformanceCounter(freq)
401 uint64_t *freq;
402 {
403 if (freq != NULL)
404 *freq = hz;
405
406 return((uint64_t)ticks);
407 }
408
409 uint8_t
410 KfRaiseIrql(irql)
411 uint8_t irql;
412 {
413 uint8_t oldirql;
414
415 oldirql = KeGetCurrentIrql();
416
417 /* I am so going to hell for this. */
418 if (oldirql > irql)
419 panic("IRQL_NOT_LESS_THAN");
420
421 if (oldirql != DISPATCH_LEVEL) {
422 sched_pin();
423 mtx_lock(&disp_lock[curthread->td_oncpu]);
424 }
425 /*printf("RAISE IRQL: %d %d\n", irql, oldirql);*/
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 uint8_t
447 KeRaiseIrqlToDpcLevel(void)
448 {
449 uint8_t irql;
450
451 KeRaiseIrql(DISPATCH_LEVEL, &irql);
452 return(irql);
453 }
454
455 static void
456 _KeLowerIrql(oldirql)
457 uint8_t oldirql;
458 {
459 KeLowerIrql(oldirql);
460 return;
461 }
462
463 static void dummy()
464 {
465 printf ("hal dummy called...\n");
466 return;
467 }
468
469 image_patch_table hal_functbl[] = {
470 IMPORT_SFUNC(KeStallExecutionProcessor, 1),
471 IMPORT_SFUNC(WRITE_PORT_ULONG, 2),
472 IMPORT_SFUNC(WRITE_PORT_USHORT, 2),
473 IMPORT_SFUNC(WRITE_PORT_UCHAR, 2),
474 IMPORT_SFUNC(WRITE_PORT_BUFFER_ULONG, 3),
475 IMPORT_SFUNC(WRITE_PORT_BUFFER_USHORT, 3),
476 IMPORT_SFUNC(WRITE_PORT_BUFFER_UCHAR, 3),
477 IMPORT_SFUNC(READ_PORT_ULONG, 1),
478 IMPORT_SFUNC(READ_PORT_USHORT, 1),
479 IMPORT_SFUNC(READ_PORT_UCHAR, 1),
480 IMPORT_SFUNC(READ_PORT_BUFFER_ULONG, 3),
481 IMPORT_SFUNC(READ_PORT_BUFFER_USHORT, 3),
482 IMPORT_SFUNC(READ_PORT_BUFFER_UCHAR, 3),
483 IMPORT_FFUNC(KfAcquireSpinLock, 1),
484 IMPORT_FFUNC(KfReleaseSpinLock, 1),
485 IMPORT_SFUNC(KeGetCurrentIrql, 0),
486 IMPORT_SFUNC(KeQueryPerformanceCounter, 1),
487 IMPORT_FFUNC(KfLowerIrql, 1),
488 IMPORT_FFUNC(KfRaiseIrql, 1),
489 IMPORT_SFUNC(KeRaiseIrqlToDpcLevel, 0),
490 #undef KeLowerIrql
491 IMPORT_SFUNC_MAP(KeLowerIrql, _KeLowerIrql, 1),
492
493 /*
494 * This last entry is a catch-all for any function we haven't
495 * implemented yet. The PE import list patching routine will
496 * use it for any function that doesn't have an explicit match
497 * in this table.
498 */
499
500 { NULL, (FUNC)dummy, NULL, 0, WINDRV_WRAP_STDCALL },
501
502 /* End of list. */
503
504 { NULL, NULL, NULL }
505 };
506 Cache object: 87b62131c8399f48cf0f9886e6823a53
|