xref: /freebsd/sys/amd64/amd64/fpu.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
1 /*-
2  * Copyright (c) 1990 William Jolitz.
3  * Copyright (c) 1991 The Regents of the University of California.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 4. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	from: @(#)npx.c	7.2 (Berkeley) 5/12/91
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/bus.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/sysctl.h>
47 #include <machine/bus.h>
48 #include <sys/rman.h>
49 #include <sys/signalvar.h>
50 
51 #include <machine/cputypes.h>
52 #include <machine/frame.h>
53 #include <machine/intr_machdep.h>
54 #include <machine/md_var.h>
55 #include <machine/pcb.h>
56 #include <machine/psl.h>
57 #include <machine/resource.h>
58 #include <machine/specialreg.h>
59 #include <machine/segments.h>
60 #include <machine/ucontext.h>
61 
62 /*
63  * Floating point support.
64  */
65 
66 #if defined(__GNUCLIKE_ASM) && !defined(lint)
67 
68 #define	fldcw(addr)		__asm("fldcw %0" : : "m" (*(addr)))
69 #define	fnclex()		__asm("fnclex")
70 #define	fninit()		__asm("fninit")
71 #define	fnstcw(addr)		__asm __volatile("fnstcw %0" : "=m" (*(addr)))
72 #define	fnstsw(addr)		__asm __volatile("fnstsw %0" : "=m" (*(addr)))
73 #define	fxrstor(addr)		__asm("fxrstor %0" : : "m" (*(addr)))
74 #define	fxsave(addr)		__asm __volatile("fxsave %0" : "=m" (*(addr)))
75 #define	ldmxcsr(r)		__asm __volatile("ldmxcsr %0" : : "m" (r))
76 #define	start_emulating()	__asm("smsw %%ax; orb %0,%%al; lmsw %%ax" \
77 				      : : "n" (CR0_TS) : "ax")
78 #define	stop_emulating()	__asm("clts")
79 
80 #else	/* !(__GNUCLIKE_ASM && !lint) */
81 
82 void	fldcw(caddr_t addr);
83 void	fnclex(void);
84 void	fninit(void);
85 void	fnstcw(caddr_t addr);
86 void	fnstsw(caddr_t addr);
87 void	fxsave(caddr_t addr);
88 void	fxrstor(caddr_t addr);
89 void	start_emulating(void);
90 void	stop_emulating(void);
91 
92 #endif	/* __GNUCLIKE_ASM && !lint */
93 
94 #define GET_FPU_CW(thread) ((thread)->td_pcb->pcb_save.sv_env.en_cw)
95 #define GET_FPU_SW(thread) ((thread)->td_pcb->pcb_save.sv_env.en_sw)
96 
97 typedef u_char bool_t;
98 
99 static	void	fpu_clean_state(void);
100 
101 int	hw_float = 1;
102 SYSCTL_INT(_hw,HW_FLOATINGPT, floatingpoint,
103 	CTLFLAG_RD, &hw_float, 0,
104 	"Floatingpoint instructions executed in hardware");
105 
106 static	struct savefpu		fpu_cleanstate;
107 static	bool_t			fpu_cleanstate_ready;
108 
109 /*
110  * Initialize floating point unit.
111  */
112 void
113 fpuinit(void)
114 {
115 	register_t savecrit;
116 	u_int mxcsr;
117 	u_short control;
118 
119 	savecrit = intr_disable();
120 	PCPU_SET(fpcurthread, 0);
121 	stop_emulating();
122 	fninit();
123 	control = __INITIAL_FPUCW__;
124 	fldcw(&control);
125 	mxcsr = __INITIAL_MXCSR__;
126 	ldmxcsr(mxcsr);
127 	fxsave(&fpu_cleanstate);
128 	if (fpu_cleanstate.sv_env.en_mxcsr_mask)
129 		cpu_mxcsr_mask = fpu_cleanstate.sv_env.en_mxcsr_mask;
130 	else
131 		cpu_mxcsr_mask = 0xFFBF;
132 	start_emulating();
133 	bzero(fpu_cleanstate.sv_fp, sizeof(fpu_cleanstate.sv_fp));
134 	bzero(fpu_cleanstate.sv_xmm, sizeof(fpu_cleanstate.sv_xmm));
135 	fpu_cleanstate_ready = 1;
136 	intr_restore(savecrit);
137 }
138 
139 /*
140  * Free coprocessor (if we have it).
141  */
142 void
143 fpuexit(struct thread *td)
144 {
145 	register_t savecrit;
146 
147 	savecrit = intr_disable();
148 	if (curthread == PCPU_GET(fpcurthread)) {
149 		stop_emulating();
150 		fxsave(&PCPU_GET(curpcb)->pcb_save);
151 		start_emulating();
152 		PCPU_SET(fpcurthread, 0);
153 	}
154 	intr_restore(savecrit);
155 }
156 
157 int
158 fpuformat()
159 {
160 
161 	return (_MC_FPFMT_XMM);
162 }
163 
164 /*
165  * The following mechanism is used to ensure that the FPE_... value
166  * that is passed as a trapcode to the signal handler of the user
167  * process does not have more than one bit set.
168  *
169  * Multiple bits may be set if the user process modifies the control
170  * word while a status word bit is already set.  While this is a sign
171  * of bad coding, we have no choise than to narrow them down to one
172  * bit, since we must not send a trapcode that is not exactly one of
173  * the FPE_ macros.
174  *
175  * The mechanism has a static table with 127 entries.  Each combination
176  * of the 7 FPU status word exception bits directly translates to a
177  * position in this table, where a single FPE_... value is stored.
178  * This FPE_... value stored there is considered the "most important"
179  * of the exception bits and will be sent as the signal code.  The
180  * precedence of the bits is based upon Intel Document "Numerical
181  * Applications", Chapter "Special Computational Situations".
182  *
183  * The macro to choose one of these values does these steps: 1) Throw
184  * away status word bits that cannot be masked.  2) Throw away the bits
185  * currently masked in the control word, assuming the user isn't
186  * interested in them anymore.  3) Reinsert status word bit 7 (stack
187  * fault) if it is set, which cannot be masked but must be presered.
188  * 4) Use the remaining bits to point into the trapcode table.
189  *
190  * The 6 maskable bits in order of their preference, as stated in the
191  * above referenced Intel manual:
192  * 1  Invalid operation (FP_X_INV)
193  * 1a   Stack underflow
194  * 1b   Stack overflow
195  * 1c   Operand of unsupported format
196  * 1d   SNaN operand.
197  * 2  QNaN operand (not an exception, irrelavant here)
198  * 3  Any other invalid-operation not mentioned above or zero divide
199  *      (FP_X_INV, FP_X_DZ)
200  * 4  Denormal operand (FP_X_DNML)
201  * 5  Numeric over/underflow (FP_X_OFL, FP_X_UFL)
202  * 6  Inexact result (FP_X_IMP)
203  */
204 static char fpetable[128] = {
205 	0,
206 	FPE_FLTINV,	/*  1 - INV */
207 	FPE_FLTUND,	/*  2 - DNML */
208 	FPE_FLTINV,	/*  3 - INV | DNML */
209 	FPE_FLTDIV,	/*  4 - DZ */
210 	FPE_FLTINV,	/*  5 - INV | DZ */
211 	FPE_FLTDIV,	/*  6 - DNML | DZ */
212 	FPE_FLTINV,	/*  7 - INV | DNML | DZ */
213 	FPE_FLTOVF,	/*  8 - OFL */
214 	FPE_FLTINV,	/*  9 - INV | OFL */
215 	FPE_FLTUND,	/*  A - DNML | OFL */
216 	FPE_FLTINV,	/*  B - INV | DNML | OFL */
217 	FPE_FLTDIV,	/*  C - DZ | OFL */
218 	FPE_FLTINV,	/*  D - INV | DZ | OFL */
219 	FPE_FLTDIV,	/*  E - DNML | DZ | OFL */
220 	FPE_FLTINV,	/*  F - INV | DNML | DZ | OFL */
221 	FPE_FLTUND,	/* 10 - UFL */
222 	FPE_FLTINV,	/* 11 - INV | UFL */
223 	FPE_FLTUND,	/* 12 - DNML | UFL */
224 	FPE_FLTINV,	/* 13 - INV | DNML | UFL */
225 	FPE_FLTDIV,	/* 14 - DZ | UFL */
226 	FPE_FLTINV,	/* 15 - INV | DZ | UFL */
227 	FPE_FLTDIV,	/* 16 - DNML | DZ | UFL */
228 	FPE_FLTINV,	/* 17 - INV | DNML | DZ | UFL */
229 	FPE_FLTOVF,	/* 18 - OFL | UFL */
230 	FPE_FLTINV,	/* 19 - INV | OFL | UFL */
231 	FPE_FLTUND,	/* 1A - DNML | OFL | UFL */
232 	FPE_FLTINV,	/* 1B - INV | DNML | OFL | UFL */
233 	FPE_FLTDIV,	/* 1C - DZ | OFL | UFL */
234 	FPE_FLTINV,	/* 1D - INV | DZ | OFL | UFL */
235 	FPE_FLTDIV,	/* 1E - DNML | DZ | OFL | UFL */
236 	FPE_FLTINV,	/* 1F - INV | DNML | DZ | OFL | UFL */
237 	FPE_FLTRES,	/* 20 - IMP */
238 	FPE_FLTINV,	/* 21 - INV | IMP */
239 	FPE_FLTUND,	/* 22 - DNML | IMP */
240 	FPE_FLTINV,	/* 23 - INV | DNML | IMP */
241 	FPE_FLTDIV,	/* 24 - DZ | IMP */
242 	FPE_FLTINV,	/* 25 - INV | DZ | IMP */
243 	FPE_FLTDIV,	/* 26 - DNML | DZ | IMP */
244 	FPE_FLTINV,	/* 27 - INV | DNML | DZ | IMP */
245 	FPE_FLTOVF,	/* 28 - OFL | IMP */
246 	FPE_FLTINV,	/* 29 - INV | OFL | IMP */
247 	FPE_FLTUND,	/* 2A - DNML | OFL | IMP */
248 	FPE_FLTINV,	/* 2B - INV | DNML | OFL | IMP */
249 	FPE_FLTDIV,	/* 2C - DZ | OFL | IMP */
250 	FPE_FLTINV,	/* 2D - INV | DZ | OFL | IMP */
251 	FPE_FLTDIV,	/* 2E - DNML | DZ | OFL | IMP */
252 	FPE_FLTINV,	/* 2F - INV | DNML | DZ | OFL | IMP */
253 	FPE_FLTUND,	/* 30 - UFL | IMP */
254 	FPE_FLTINV,	/* 31 - INV | UFL | IMP */
255 	FPE_FLTUND,	/* 32 - DNML | UFL | IMP */
256 	FPE_FLTINV,	/* 33 - INV | DNML | UFL | IMP */
257 	FPE_FLTDIV,	/* 34 - DZ | UFL | IMP */
258 	FPE_FLTINV,	/* 35 - INV | DZ | UFL | IMP */
259 	FPE_FLTDIV,	/* 36 - DNML | DZ | UFL | IMP */
260 	FPE_FLTINV,	/* 37 - INV | DNML | DZ | UFL | IMP */
261 	FPE_FLTOVF,	/* 38 - OFL | UFL | IMP */
262 	FPE_FLTINV,	/* 39 - INV | OFL | UFL | IMP */
263 	FPE_FLTUND,	/* 3A - DNML | OFL | UFL | IMP */
264 	FPE_FLTINV,	/* 3B - INV | DNML | OFL | UFL | IMP */
265 	FPE_FLTDIV,	/* 3C - DZ | OFL | UFL | IMP */
266 	FPE_FLTINV,	/* 3D - INV | DZ | OFL | UFL | IMP */
267 	FPE_FLTDIV,	/* 3E - DNML | DZ | OFL | UFL | IMP */
268 	FPE_FLTINV,	/* 3F - INV | DNML | DZ | OFL | UFL | IMP */
269 	FPE_FLTSUB,	/* 40 - STK */
270 	FPE_FLTSUB,	/* 41 - INV | STK */
271 	FPE_FLTUND,	/* 42 - DNML | STK */
272 	FPE_FLTSUB,	/* 43 - INV | DNML | STK */
273 	FPE_FLTDIV,	/* 44 - DZ | STK */
274 	FPE_FLTSUB,	/* 45 - INV | DZ | STK */
275 	FPE_FLTDIV,	/* 46 - DNML | DZ | STK */
276 	FPE_FLTSUB,	/* 47 - INV | DNML | DZ | STK */
277 	FPE_FLTOVF,	/* 48 - OFL | STK */
278 	FPE_FLTSUB,	/* 49 - INV | OFL | STK */
279 	FPE_FLTUND,	/* 4A - DNML | OFL | STK */
280 	FPE_FLTSUB,	/* 4B - INV | DNML | OFL | STK */
281 	FPE_FLTDIV,	/* 4C - DZ | OFL | STK */
282 	FPE_FLTSUB,	/* 4D - INV | DZ | OFL | STK */
283 	FPE_FLTDIV,	/* 4E - DNML | DZ | OFL | STK */
284 	FPE_FLTSUB,	/* 4F - INV | DNML | DZ | OFL | STK */
285 	FPE_FLTUND,	/* 50 - UFL | STK */
286 	FPE_FLTSUB,	/* 51 - INV | UFL | STK */
287 	FPE_FLTUND,	/* 52 - DNML | UFL | STK */
288 	FPE_FLTSUB,	/* 53 - INV | DNML | UFL | STK */
289 	FPE_FLTDIV,	/* 54 - DZ | UFL | STK */
290 	FPE_FLTSUB,	/* 55 - INV | DZ | UFL | STK */
291 	FPE_FLTDIV,	/* 56 - DNML | DZ | UFL | STK */
292 	FPE_FLTSUB,	/* 57 - INV | DNML | DZ | UFL | STK */
293 	FPE_FLTOVF,	/* 58 - OFL | UFL | STK */
294 	FPE_FLTSUB,	/* 59 - INV | OFL | UFL | STK */
295 	FPE_FLTUND,	/* 5A - DNML | OFL | UFL | STK */
296 	FPE_FLTSUB,	/* 5B - INV | DNML | OFL | UFL | STK */
297 	FPE_FLTDIV,	/* 5C - DZ | OFL | UFL | STK */
298 	FPE_FLTSUB,	/* 5D - INV | DZ | OFL | UFL | STK */
299 	FPE_FLTDIV,	/* 5E - DNML | DZ | OFL | UFL | STK */
300 	FPE_FLTSUB,	/* 5F - INV | DNML | DZ | OFL | UFL | STK */
301 	FPE_FLTRES,	/* 60 - IMP | STK */
302 	FPE_FLTSUB,	/* 61 - INV | IMP | STK */
303 	FPE_FLTUND,	/* 62 - DNML | IMP | STK */
304 	FPE_FLTSUB,	/* 63 - INV | DNML | IMP | STK */
305 	FPE_FLTDIV,	/* 64 - DZ | IMP | STK */
306 	FPE_FLTSUB,	/* 65 - INV | DZ | IMP | STK */
307 	FPE_FLTDIV,	/* 66 - DNML | DZ | IMP | STK */
308 	FPE_FLTSUB,	/* 67 - INV | DNML | DZ | IMP | STK */
309 	FPE_FLTOVF,	/* 68 - OFL | IMP | STK */
310 	FPE_FLTSUB,	/* 69 - INV | OFL | IMP | STK */
311 	FPE_FLTUND,	/* 6A - DNML | OFL | IMP | STK */
312 	FPE_FLTSUB,	/* 6B - INV | DNML | OFL | IMP | STK */
313 	FPE_FLTDIV,	/* 6C - DZ | OFL | IMP | STK */
314 	FPE_FLTSUB,	/* 6D - INV | DZ | OFL | IMP | STK */
315 	FPE_FLTDIV,	/* 6E - DNML | DZ | OFL | IMP | STK */
316 	FPE_FLTSUB,	/* 6F - INV | DNML | DZ | OFL | IMP | STK */
317 	FPE_FLTUND,	/* 70 - UFL | IMP | STK */
318 	FPE_FLTSUB,	/* 71 - INV | UFL | IMP | STK */
319 	FPE_FLTUND,	/* 72 - DNML | UFL | IMP | STK */
320 	FPE_FLTSUB,	/* 73 - INV | DNML | UFL | IMP | STK */
321 	FPE_FLTDIV,	/* 74 - DZ | UFL | IMP | STK */
322 	FPE_FLTSUB,	/* 75 - INV | DZ | UFL | IMP | STK */
323 	FPE_FLTDIV,	/* 76 - DNML | DZ | UFL | IMP | STK */
324 	FPE_FLTSUB,	/* 77 - INV | DNML | DZ | UFL | IMP | STK */
325 	FPE_FLTOVF,	/* 78 - OFL | UFL | IMP | STK */
326 	FPE_FLTSUB,	/* 79 - INV | OFL | UFL | IMP | STK */
327 	FPE_FLTUND,	/* 7A - DNML | OFL | UFL | IMP | STK */
328 	FPE_FLTSUB,	/* 7B - INV | DNML | OFL | UFL | IMP | STK */
329 	FPE_FLTDIV,	/* 7C - DZ | OFL | UFL | IMP | STK */
330 	FPE_FLTSUB,	/* 7D - INV | DZ | OFL | UFL | IMP | STK */
331 	FPE_FLTDIV,	/* 7E - DNML | DZ | OFL | UFL | IMP | STK */
332 	FPE_FLTSUB,	/* 7F - INV | DNML | DZ | OFL | UFL | IMP | STK */
333 };
334 
335 /*
336  * Preserve the FP status word, clear FP exceptions, then generate a SIGFPE.
337  *
338  * Clearing exceptions is necessary mainly to avoid IRQ13 bugs.  We now
339  * depend on longjmp() restoring a usable state.  Restoring the state
340  * or examining it might fail if we didn't clear exceptions.
341  *
342  * The error code chosen will be one of the FPE_... macros. It will be
343  * sent as the second argument to old BSD-style signal handlers and as
344  * "siginfo_t->si_code" (second argument) to SA_SIGINFO signal handlers.
345  *
346  * XXX the FP state is not preserved across signal handlers.  So signal
347  * handlers cannot afford to do FP unless they preserve the state or
348  * longjmp() out.  Both preserving the state and longjmp()ing may be
349  * destroyed by IRQ13 bugs.  Clearing FP exceptions is not an acceptable
350  * solution for signals other than SIGFPE.
351  */
352 int
353 fputrap()
354 {
355 	register_t savecrit;
356 	u_short control, status;
357 
358 	savecrit = intr_disable();
359 
360 	/*
361 	 * Interrupt handling (for another interrupt) may have pushed the
362 	 * state to memory.  Fetch the relevant parts of the state from
363 	 * wherever they are.
364 	 */
365 	if (PCPU_GET(fpcurthread) != curthread) {
366 		control = GET_FPU_CW(curthread);
367 		status = GET_FPU_SW(curthread);
368 	} else {
369 		fnstcw(&control);
370 		fnstsw(&status);
371 	}
372 
373 	if (PCPU_GET(fpcurthread) == curthread)
374 		fnclex();
375 	intr_restore(savecrit);
376 	return (fpetable[status & ((~control & 0x3f) | 0x40)]);
377 }
378 
379 /*
380  * Implement device not available (DNA) exception
381  *
382  * It would be better to switch FP context here (if curthread != fpcurthread)
383  * and not necessarily for every context switch, but it is too hard to
384  * access foreign pcb's.
385  */
386 
387 static int err_count = 0;
388 
389 int
390 fpudna()
391 {
392 	struct pcb *pcb;
393 	register_t s;
394 
395 	if (PCPU_GET(fpcurthread) == curthread) {
396 		printf("fpudna: fpcurthread == curthread %d times\n",
397 		    ++err_count);
398 		stop_emulating();
399 		return (1);
400 	}
401 	if (PCPU_GET(fpcurthread) != NULL) {
402 		printf("fpudna: fpcurthread = %p (%d), curthread = %p (%d)\n",
403 		       PCPU_GET(fpcurthread),
404 		       PCPU_GET(fpcurthread)->td_proc->p_pid,
405 		       curthread, curthread->td_proc->p_pid);
406 		panic("fpudna");
407 	}
408 	s = intr_disable();
409 	stop_emulating();
410 	/*
411 	 * Record new context early in case frstor causes a trap.
412 	 */
413 	PCPU_SET(fpcurthread, curthread);
414 	pcb = PCPU_GET(curpcb);
415 
416 	fpu_clean_state();
417 
418 	if ((pcb->pcb_flags & PCB_FPUINITDONE) == 0) {
419 		/*
420 		 * This is the first time this thread has used the FPU,
421 		 * explicitly load sanitized registers.
422 		 */
423 		fxrstor(&fpu_cleanstate);
424 		pcb->pcb_flags |= PCB_FPUINITDONE;
425 	} else
426 		fxrstor(&pcb->pcb_save);
427 	intr_restore(s);
428 
429 	return (1);
430 }
431 
432 /*
433  * This should be called with interrupts disabled and only when the owning
434  * FPU thread is non-null.
435  */
436 void
437 fpudrop()
438 {
439 	struct thread *td;
440 
441 	td = PCPU_GET(fpcurthread);
442 	PCPU_SET(fpcurthread, NULL);
443 	td->td_pcb->pcb_flags &= ~PCB_FPUINITDONE;
444 	start_emulating();
445 }
446 
447 /*
448  * Get the state of the FPU without dropping ownership (if possible).
449  * It returns the FPU ownership status.
450  */
451 int
452 fpugetregs(struct thread *td, struct savefpu *addr)
453 {
454 	register_t s;
455 
456 	if ((td->td_pcb->pcb_flags & PCB_FPUINITDONE) == 0) {
457 		if (fpu_cleanstate_ready)
458 			bcopy(&fpu_cleanstate, addr, sizeof(fpu_cleanstate));
459 		else
460 			bzero(addr, sizeof(*addr));
461 		return (_MC_FPOWNED_NONE);
462 	}
463 	s = intr_disable();
464 	if (td == PCPU_GET(fpcurthread)) {
465 		fxsave(addr);
466 		intr_restore(s);
467 		return (_MC_FPOWNED_FPU);
468 	} else {
469 		intr_restore(s);
470 		bcopy(&td->td_pcb->pcb_save, addr, sizeof(*addr));
471 		return (_MC_FPOWNED_PCB);
472 	}
473 }
474 
475 /*
476  * Set the state of the FPU.
477  */
478 void
479 fpusetregs(struct thread *td, struct savefpu *addr)
480 {
481 	register_t s;
482 
483 	s = intr_disable();
484 	if (td == PCPU_GET(fpcurthread)) {
485 		fpu_clean_state();
486 		fxrstor(addr);
487 		intr_restore(s);
488 	} else {
489 		intr_restore(s);
490 		bcopy(addr, &td->td_pcb->pcb_save, sizeof(*addr));
491 	}
492 	curthread->td_pcb->pcb_flags |= PCB_FPUINITDONE;
493 }
494 
495 /*
496  * On AuthenticAMD processors, the fxrstor instruction does not restore
497  * the x87's stored last instruction pointer, last data pointer, and last
498  * opcode values, except in the rare case in which the exception summary
499  * (ES) bit in the x87 status word is set to 1.
500  *
501  * In order to avoid leaking this information across processes, we clean
502  * these values by performing a dummy load before executing fxrstor().
503  */
504 static	double	dummy_variable = 0.0;
505 static void
506 fpu_clean_state(void)
507 {
508 	u_short status;
509 
510 	/*
511 	 * Clear the ES bit in the x87 status word if it is currently
512 	 * set, in order to avoid causing a fault in the upcoming load.
513 	 */
514 	fnstsw(&status);
515 	if (status & 0x80)
516 		fnclex();
517 
518 	/*
519 	 * Load the dummy variable into the x87 stack.  This mangles
520 	 * the x87 stack, but we don't care since we're about to call
521 	 * fxrstor() anyway.
522 	 */
523 	__asm __volatile("ffree %%st(7); fld %0" : : "m" (dummy_variable));
524 }
525 
526 /*
527  * This really sucks.  We want the acpi version only, but it requires
528  * the isa_if.h file in order to get the definitions.
529  */
530 #include "opt_isa.h"
531 #ifdef DEV_ISA
532 #include <isa/isavar.h>
533 /*
534  * This sucks up the legacy ISA support assignments from PNPBIOS/ACPI.
535  */
536 static struct isa_pnp_id fpupnp_ids[] = {
537 	{ 0x040cd041, "Legacy ISA coprocessor support" }, /* PNP0C04 */
538 	{ 0 }
539 };
540 
541 static int
542 fpupnp_probe(device_t dev)
543 {
544 	int result;
545 
546 	result = ISA_PNP_PROBE(device_get_parent(dev), dev, fpupnp_ids);
547 	if (result <= 0)
548 		device_quiet(dev);
549 	return (result);
550 }
551 
552 static int
553 fpupnp_attach(device_t dev)
554 {
555 
556 	return (0);
557 }
558 
559 static device_method_t fpupnp_methods[] = {
560 	/* Device interface */
561 	DEVMETHOD(device_probe,		fpupnp_probe),
562 	DEVMETHOD(device_attach,	fpupnp_attach),
563 	DEVMETHOD(device_detach,	bus_generic_detach),
564 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
565 	DEVMETHOD(device_suspend,	bus_generic_suspend),
566 	DEVMETHOD(device_resume,	bus_generic_resume),
567 
568 	{ 0, 0 }
569 };
570 
571 static driver_t fpupnp_driver = {
572 	"fpupnp",
573 	fpupnp_methods,
574 	1,			/* no softc */
575 };
576 
577 static devclass_t fpupnp_devclass;
578 
579 DRIVER_MODULE(fpupnp, acpi, fpupnp_driver, fpupnp_devclass, 0, 0);
580 #endif	/* DEV_ISA */
581