xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision b7ae68fde0d8ef1c03714e8bbb1ee7c6118ea93b)
1 /*	$NetBSD: linux_machdep.c,v 1.15 2006/09/20 09:54:55 manu Exp $ */
2 
3 /*-
4  * Copyright (c) 2005 Emmanuel Dreyfus, 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  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by Emmanuel Dreyfus
17  * 4. The name of the author may not be used to endorse or promote
18  *    products derived from this software without specific prior written
19  *    permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE THE AUTHOR AND CONTRIBUTORS ``AS IS''
22  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
23  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
24  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <sys/cdefs.h>
35 
36 __KERNEL_RCSID(0, "$NetBSD: linux_machdep.c,v 1.15 2006/09/20 09:54:55 manu Exp $");
37 
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/systm.h>
41 #include <sys/signal.h>
42 #include <sys/exec.h>
43 #include <sys/proc.h>
44 #include <sys/ptrace.h> /* for process_read_fpregs() */
45 #include <sys/user.h>
46 #include <sys/wait.h>
47 #include <sys/ucontext.h>
48 
49 #include <machine/reg.h>
50 #include <machine/pcb.h>
51 #include <machine/fpu.h>
52 #include <machine/mcontext.h>
53 #include <machine/specialreg.h>
54 #include <machine/vmparam.h>
55 
56 #include <compat/linux/common/linux_signal.h>
57 #include <compat/linux/common/linux_errno.h>
58 #include <compat/linux/common/linux_exec.h>
59 #include <compat/linux/common/linux_ioctl.h>
60 #include <compat/linux/common/linux_prctl.h>
61 #include <compat/linux/common/linux_machdep.h>
62 #include <compat/linux/linux_syscall.h>
63 #include <compat/linux/linux_syscallargs.h>
64 
65 static void linux_buildcontext(struct lwp *, void *, void *);
66 
67 void
68 linux_setregs(l, epp, stack)
69         struct lwp *l;
70 	struct exec_package *epp;
71 	u_long stack;
72 {
73 	struct pcb *pcb = &l->l_addr->u_pcb;
74 	struct trapframe *tf;
75 
76 	/* If we were using the FPU, forget about it. */
77 	if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
78 		fpusave_lwp(l, 0);
79 
80 	l->l_md.md_flags &= ~MDP_USEDFPU;
81 	pcb->pcb_flags = 0;
82 	pcb->pcb_savefpu.fp_fxsave.fx_fcw = __NetBSD_NPXCW__;
83 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr = __INITIAL_MXCSR__;
84 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr_mask = __INITIAL_MXCSR_MASK__;
85 	pcb->pcb_fs = 0;
86 	pcb->pcb_gs = 0;
87 
88 	l->l_proc->p_flag &= ~P_32;
89 
90 	tf = l->l_md.md_regs;
91 	tf->tf_rax = 0;
92 	tf->tf_rbx = 0;
93 	tf->tf_rcx = epp->ep_entry;
94 	tf->tf_rdx = 0;
95 	tf->tf_rsi = 0;
96 	tf->tf_rdi = 0;
97 	tf->tf_rbp = 0;
98 	tf->tf_rsp = stack;
99 	tf->tf_r8 = 0;
100 	tf->tf_r9 = 0;
101 	tf->tf_r10 = 0;
102 	tf->tf_r11 = 0;
103 	tf->tf_r12 = 0;
104 	tf->tf_r13 = 0;
105 	tf->tf_r14 = 0;
106 	tf->tf_r15 = 0;
107 	tf->tf_rip = epp->ep_entry;
108 	tf->tf_rflags = PSL_USERSET;
109 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
110 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
111 	tf->tf_ds = 0;
112 	tf->tf_es = 0;
113 	tf->tf_fs = 0;
114 	tf->tf_gs = 0;
115 
116 	return;
117 }
118 
119 void
120 linux_sendsig(ksi, mask)
121 	const ksiginfo_t *ksi;
122 	const sigset_t *mask;
123 {
124 	struct lwp *l = curlwp;
125 	struct proc *p = l->l_proc;
126 	struct sigacts *ps = p->p_sigacts;
127 	int onstack;
128 	int sig = ksi->ksi_signo;
129 	struct linux_rt_sigframe *sfp, sigframe;
130 	struct linux__fpstate *fpsp, fpstate;
131 	struct fpreg fpregs;
132 	struct trapframe *tf = l->l_md.md_regs;
133 	sig_t catcher = SIGACTION(p, sig).sa_handler;
134 	linux_sigset_t lmask;
135 	char *sp;
136 	int error;
137 
138 	/* Do we need to jump onto the signal stack? */
139 	onstack =
140 	    (p->p_sigctx.ps_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
141 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
142 
143 	/* Allocate space for the signal handler context. */
144 	if (onstack)
145 		sp = ((caddr_t)p->p_sigctx.ps_sigstk.ss_sp +
146 		    p->p_sigctx.ps_sigstk.ss_size);
147 	else
148 		sp = (caddr_t)tf->tf_rsp - 128;
149 
150 
151 	/*
152 	 * Save FPU state, if any
153 	 */
154 	if (l->l_md.md_flags & MDP_USEDFPU) {
155 		sp = (char *)
156 		    (((long)sp - sizeof(struct linux__fpstate)) & ~0xfUL);
157 		fpsp = (struct linux__fpstate *)sp;
158 
159 		(void)process_read_fpregs(l, &fpregs);
160 		bzero(&fpstate, sizeof(fpstate));
161 
162 		fpstate.cwd = fpregs.fp_fcw;
163 		fpstate.swd = fpregs.fp_fsw;
164 		fpstate.twd = fpregs.fp_ftw;
165 		fpstate.fop = fpregs.fp_fop;
166 		fpstate.rip = fpregs.fp_rip;
167 		fpstate.rdp = fpregs.fp_rdp;
168 		fpstate.mxcsr = fpregs.fp_mxcsr;
169 		fpstate.mxcsr_mask = fpregs.fp_mxcsr_mask;
170 		memcpy(&fpstate.st_space, &fpregs.fp_st,
171 		    sizeof(fpstate.st_space));
172 		memcpy(&fpstate.xmm_space, &fpregs.fp_xmm,
173 		    sizeof(fpstate.xmm_space));
174 
175 		if ((error = copyout(&fpstate, fpsp, sizeof(fpstate))) != 0) {
176 			sigexit(l, SIGILL);
177 			return;
178 		}
179 	} else {
180 		fpsp = NULL;
181 	}
182 
183 	/*
184 	 * Populate the rt_sigframe
185 	 */
186 	sp = (char *)
187 	    ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8);
188 	sfp = (struct linux_rt_sigframe *)sp;
189 
190 	bzero(&sigframe, sizeof(sigframe));
191 	if (ps->sa_sigdesc[sig].sd_vers != 0)
192 		sigframe.pretcode =
193 		    (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp;
194 	else
195 		sigframe.pretcode = NULL;
196 
197 	/*
198 	 * The user context
199 	 */
200 	sigframe.uc.luc_flags = 0;
201 	sigframe.uc.luc_link = NULL;
202 
203 	/* This is used regardless of SA_ONSTACK in Linux */
204 	sigframe.uc.luc_stack.ss_sp = p->p_sigctx.ps_sigstk.ss_sp;
205 	sigframe.uc.luc_stack.ss_size = p->p_sigctx.ps_sigstk.ss_size;
206 	sigframe.uc.luc_stack.ss_flags = 0;
207 	if (p->p_sigctx.ps_sigstk.ss_flags & SS_ONSTACK)
208 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK;
209 	if (p->p_sigctx.ps_sigstk.ss_flags & SS_DISABLE)
210 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE;
211 
212 	sigframe.uc.luc_mcontext.r8 = tf->tf_r8;
213 	sigframe.uc.luc_mcontext.r9 = tf->tf_r9;
214 	sigframe.uc.luc_mcontext.r10 = tf->tf_r10;
215 	sigframe.uc.luc_mcontext.r11 = tf->tf_r11;
216 	sigframe.uc.luc_mcontext.r12 = tf->tf_r12;
217 	sigframe.uc.luc_mcontext.r13 = tf->tf_r13;
218 	sigframe.uc.luc_mcontext.r14 = tf->tf_r14;
219 	sigframe.uc.luc_mcontext.r15 = tf->tf_r15;
220 	sigframe.uc.luc_mcontext.rdi = tf->tf_rdi;
221 	sigframe.uc.luc_mcontext.rsi = tf->tf_rsi;
222 	sigframe.uc.luc_mcontext.rbp = tf->tf_rbp;
223 	sigframe.uc.luc_mcontext.rbx = tf->tf_rbx;
224 	sigframe.uc.luc_mcontext.rdx = tf->tf_rdx;
225 	sigframe.uc.luc_mcontext.rcx = tf->tf_rcx;
226 	sigframe.uc.luc_mcontext.rsp = tf->tf_rsp;
227 	sigframe.uc.luc_mcontext.rip = tf->tf_rip;
228 	sigframe.uc.luc_mcontext.eflags = tf->tf_rflags;
229 	sigframe.uc.luc_mcontext.cs = tf->tf_cs;
230 	sigframe.uc.luc_mcontext.gs = tf->tf_gs;
231 	sigframe.uc.luc_mcontext.fs = tf->tf_fs;
232 	sigframe.uc.luc_mcontext.err = tf->tf_err;
233 	sigframe.uc.luc_mcontext.trapno = tf->tf_trapno;
234 	native_to_linux_sigset(&lmask, mask);
235 	sigframe.uc.luc_mcontext.oldmask = lmask.sig[0];
236 	sigframe.uc.luc_mcontext.cr2 = (long)l->l_addr->u_pcb.pcb_onfault;
237 	sigframe.uc.luc_mcontext.fpstate = fpsp;
238 	native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask);
239 
240 	/*
241 	 * the siginfo structure
242 	 */
243 	sigframe.info.lsi_signo = native_to_linux_signo[sig];
244 	sigframe.info.lsi_errno = native_to_linux_errno[ksi->ksi_errno];
245 	sigframe.info.lsi_code = ksi->ksi_code;
246 
247 	/* XXX This is a rought conversion, taken from i386 code */
248 	switch (sigframe.info.lsi_signo) {
249 	case LINUX_SIGILL:
250 	case LINUX_SIGFPE:
251 	case LINUX_SIGSEGV:
252 	case LINUX_SIGBUS:
253 	case LINUX_SIGTRAP:
254 		sigframe.info._sifields._sigfault._addr = ksi->ksi_addr;
255 		break;
256 	case LINUX_SIGCHLD:
257 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
258 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
259 		sigframe.info._sifields._sigchld._utime = ksi->ksi_utime;
260 		sigframe.info._sifields._sigchld._stime = ksi->ksi_stime;
261 
262 		if (WCOREDUMP(ksi->ksi_status)) {
263 			sigframe.info.lsi_code = LINUX_CLD_DUMPED;
264 			sigframe.info._sifields._sigchld._status =
265 			    _WSTATUS(ksi->ksi_status);
266 		} else if (_WSTATUS(ksi->ksi_status)) {
267 			sigframe.info.lsi_code = LINUX_CLD_KILLED;
268 			sigframe.info._sifields._sigchld._status =
269 			    _WSTATUS(ksi->ksi_status);
270 		} else {
271 			sigframe.info.lsi_code = LINUX_CLD_EXITED;
272 			sigframe.info._sifields._sigchld._status =
273 			    ((ksi->ksi_status & 0xff00U) >> 8);
274 		}
275 		break;
276 	case LINUX_SIGIO:
277 		sigframe.info._sifields._sigpoll._band = ksi->ksi_band;
278 		sigframe.info._sifields._sigpoll._fd = ksi->ksi_fd;
279 		break;
280 	default:
281 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
282 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
283 		if ((sigframe.info.lsi_signo == LINUX_SIGALRM) ||
284 		    (sigframe.info.lsi_signo >= LINUX_SIGRTMIN))
285 			sigframe.info._sifields._timer._sigval.sival_ptr =
286 			     ksi->ksi_sigval.sival_ptr;
287 		break;
288 	}
289 
290 	if ((error = copyout(&sigframe, sp, sizeof(sigframe))) != 0) {
291 		sigexit(l, SIGILL);
292 		return;
293 	}
294 
295 	linux_buildcontext(l, catcher, sp);
296 	tf->tf_rdi = sigframe.info.lsi_signo;
297 	tf->tf_rax = 0;
298 	tf->tf_rsi = (long)&sfp->info;
299 	tf->tf_rdx = (long)&sfp->uc;
300 
301 	/*
302 	 * Remember we use signal stack
303 	 */
304 	if (onstack)
305 		p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK;
306 	return;
307 }
308 
309 int
310 linux_sys_modify_ldt(l, v, retval)
311         struct lwp *l;
312         void *v;
313         register_t *retval;
314 {
315 	printf("linux_sys_modify_ldt\n");
316 	return 0;
317 }
318 
319 int
320 linux_sys_iopl(l, v, retval)
321         struct lwp *l;
322         void *v;
323         register_t *retval;
324 {
325 	return 0;
326 }
327 
328 int
329 linux_sys_ioperm(l, v, retval)
330         struct lwp *l;
331         void *v;
332         register_t *retval;
333 {
334 	return 0;
335 }
336 
337 dev_t
338 linux_fakedev(dev, raw)
339         dev_t dev;
340 	int raw;
341 {
342 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
343 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
344 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
345 }
346 
347 int
348 linux_machdepioctl(l, v, retval)
349         struct lwp *l;
350         void *v;
351         register_t *retval;
352 {
353 	return 0;
354 }
355 
356 int
357 linux_sys_rt_sigreturn(l, v, retval)
358         struct lwp *l;
359         void *v;
360         register_t *retval;
361 {
362 	struct linux_ucontext *luctx;
363 	struct trapframe *tf = l->l_md.md_regs;
364 	struct linux_sigcontext *lsigctx;
365 	struct linux__fpstate fpstate;
366 	struct linux_rt_sigframe frame, *fp;
367 	ucontext_t uctx;
368 	mcontext_t *mctx;
369 	struct fxsave64 *fxsave;
370 	int error;
371 
372 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
373 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
374 		sigexit(l, SIGILL);
375 		return error;
376 	}
377 	luctx = &frame.uc;
378 	lsigctx = &luctx->luc_mcontext;
379 
380 	bzero(&uctx, sizeof(uctx));
381 	mctx = (mcontext_t *)&uctx.uc_mcontext;
382 	fxsave = (struct fxsave64 *)&mctx->__fpregs;
383 
384 	/*
385 	 * Set the flags. Linux always have CPU, stack and signal state,
386 	 * FPU is optional. uc_flags is not used to tell what we have.
387 	 */
388 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
389 	if (lsigctx->fpstate != NULL)
390 		uctx.uc_flags |= _UC_FPU;
391 	uctx.uc_link = NULL;
392 
393 	/*
394 	 * Signal set
395 	 */
396 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
397 
398 	/*
399 	 * CPU state
400 	 */
401 	mctx->__gregs[_REG_R8] = lsigctx->r8;
402 	mctx->__gregs[_REG_R9] = lsigctx->r9;
403 	mctx->__gregs[_REG_R10] = lsigctx->r10;
404 	mctx->__gregs[_REG_R11] = lsigctx->r11;
405 	mctx->__gregs[_REG_R12] = lsigctx->r12;
406 	mctx->__gregs[_REG_R13] = lsigctx->r13;
407 	mctx->__gregs[_REG_R14] = lsigctx->r14;
408 	mctx->__gregs[_REG_R15] = lsigctx->r15;
409 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
410 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
411 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
412 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
413 	mctx->__gregs[_REG_RAX] = tf->tf_rax;
414 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
415 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
416 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
417 	mctx->__gregs[_REG_RFL] = lsigctx->eflags;
418 	mctx->__gregs[_REG_CS] = lsigctx->cs;
419 	mctx->__gregs[_REG_GS] = lsigctx->gs;
420 	mctx->__gregs[_REG_FS] = lsigctx->fs;
421 	mctx->__gregs[_REG_ERR] = lsigctx->err;
422 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
423 	mctx->__gregs[_REG_ES] = tf->tf_es;
424 	mctx->__gregs[_REG_DS] = tf->tf_ds;
425 	mctx->__gregs[_REG_URSP] = lsigctx->rsp; /* XXX */
426 	mctx->__gregs[_REG_SS] = tf->tf_ss;
427 
428 	/*
429 	 * FPU state
430 	 */
431 	if (lsigctx->fpstate != NULL) {
432 		error = copyin(lsigctx->fpstate, &fpstate, sizeof(fpstate));
433 		if (error != 0) {
434 			sigexit(l, SIGILL);
435 			return error;
436 		}
437 
438 		fxsave->fx_fcw = fpstate.cwd;
439 		fxsave->fx_fsw = fpstate.swd;
440 		fxsave->fx_ftw = fpstate.twd;
441 		fxsave->fx_fop = fpstate.fop;
442 		fxsave->fx_rip = fpstate.rip;
443 		fxsave->fx_rdp = fpstate.rdp;
444 		fxsave->fx_mxcsr = fpstate.mxcsr;
445 		fxsave->fx_mxcsr_mask = fpstate.mxcsr_mask;
446 		memcpy(&fxsave->fx_st, &fpstate.st_space,
447 		    sizeof(fxsave->fx_st));
448 		memcpy(&fxsave->fx_xmm, &fpstate.xmm_space,
449 		    sizeof(fxsave->fx_xmm));
450 	}
451 
452 	/*
453 	 * And the stack
454 	 */
455 	uctx.uc_stack.ss_flags = 0;
456 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK);
457 		uctx.uc_stack.ss_flags = SS_ONSTACK;
458 
459 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE);
460 		uctx.uc_stack.ss_flags = SS_DISABLE;
461 
462 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
463 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
464 
465 	/*
466 	 * And let setucontext deal with that.
467 	 */
468 	return setucontext(l, &uctx);
469 }
470 
471 int
472 linux_sys_arch_prctl(l, v, retval)
473 	struct lwp *l;
474 	void *v;
475 	register_t *retval;
476 {
477 	struct linux_sys_arch_prctl_args /* {
478 		syscallarg(int) code;
479 		syscallarg(unsigned long) addr;
480 	} */ *uap = v;
481 	struct pcb *pcb = &l->l_addr->u_pcb;
482 	struct trapframe *tf = l->l_md.md_regs;
483 	int error;
484 	uint64_t taddr;
485 
486 	switch(SCARG(uap, code)) {
487 	case LINUX_ARCH_SET_GS:
488 		taddr = SCARG(uap, addr);
489 		if (taddr >= VM_MAXUSER_ADDRESS)
490 			return EINVAL;
491 		pcb->pcb_gs = taddr;
492 		pcb->pcb_flags |= PCB_GS64;
493 		if (l == curlwp)
494 			wrmsr(MSR_KERNELGSBASE, taddr);
495 		break;
496 
497 	case LINUX_ARCH_GET_GS:
498 		if (pcb->pcb_flags & PCB_GS64)
499 			taddr = pcb->pcb_gs;
500 		else {
501 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
502 			if (error != 0)
503 				return error;
504 		}
505 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
506 		if (error != 0)
507 			return error;
508 		break;
509 
510 	case LINUX_ARCH_SET_FS:
511 		taddr = SCARG(uap, addr);
512 		if (taddr >= VM_MAXUSER_ADDRESS)
513 			return EINVAL;
514 		pcb->pcb_fs = taddr;
515 		pcb->pcb_flags |= PCB_FS64;
516 		if (l == curlwp)
517 			wrmsr(MSR_FSBASE, taddr);
518 		break;
519 
520 	case LINUX_ARCH_GET_FS:
521 		if (pcb->pcb_flags & PCB_FS64)
522 			taddr = pcb->pcb_fs;
523 		else {
524 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
525 			if (error != 0)
526 				return error;
527 		}
528 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
529 		if (error != 0)
530 			return error;
531 		break;
532 
533 	default:
534 #ifdef DEBUG_LINUX
535 		printf("linux_sys_arch_prctl: unexpected code %d\n",
536 		    SCARG(uap, code));
537 #endif
538 		return EINVAL;
539 	}
540 
541 	return 0;
542 }
543 
544 const int linux_vsyscall_to_syscall[] = {
545 	LINUX_SYS_gettimeofday,
546 	LINUX_SYS_time,
547 	LINUX_SYS_nosys,
548 	LINUX_SYS_nosys,
549 };
550 
551 int
552 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
553 {
554 	struct trapframe *tf = arg;
555 	uint64_t retaddr;
556 	int vsyscallnr;
557 
558 	/*
559 	 * Check for a vsyscall. %rip must be the fault address,
560 	 * and the address must be in the Linux vsyscall area.
561 	 * Also, vsyscalls are only done at 1024-byte boundaries.
562 	 */
563 
564 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
565 		return 0;
566 
567 	if (trapaddr != tf->tf_rip)
568 		return 0;
569 
570 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
571 		return 0;
572 
573 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
574 
575 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
576 		return 0;
577 
578 	/*
579 	 * Get the return address from the top of the stack,
580 	 * and fix up the return address.
581 	 * This assumes the faulting instruction was callq *reg,
582 	 * which is the only way that vsyscalls are ever entered.
583 	 */
584 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
585 		return 0;
586 	tf->tf_rip = retaddr;
587 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
588 	tf->tf_rsp += 8;	/* "pop" the return address */
589 
590 #if 0
591 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
592 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
593 	    vsyscallnr, (int)tf->tf_rax);
594 #endif
595 
596 	(*l->l_proc->p_md.md_syscall)(tf);
597 
598 	return 1;
599 }
600 
601 static void
602 linux_buildcontext(struct lwp *l, void *catcher, void *f)
603 {
604 	struct trapframe *tf = l->l_md.md_regs;
605 
606 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
607 	tf->tf_rip = (u_int64_t)catcher;
608 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
609 	tf->tf_rflags &= ~(PSL_T|PSL_VM|PSL_AC);
610 	tf->tf_rsp = (u_int64_t)f;
611 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
612 }
613 
614 void *
615 linux_get_newtls(l)
616 	struct lwp *l;
617 {
618 	struct trapframe *tf = l->l_md.md_regs;
619 
620 	return (void *)tf->tf_r8;
621 }
622 
623 int
624 linux_set_newtls(l, tls)
625 	struct lwp *l;
626 	void *tls;
627 {
628 	struct linux_sys_arch_prctl_args cup;
629 	register_t retval;
630 
631 	SCARG(&cup, code) = LINUX_ARCH_SET_FS;
632 	SCARG(&cup, addr) = (unsigned long)tls;
633 
634 	return linux_sys_arch_prctl(l, &cup, &retval);
635 }
636