xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision b5677b36047b601b9addaaa494a58ceae82c2a6c)
1 /*	$NetBSD: linux_machdep.c,v 1.35 2009/03/18 16:00:16 cegger 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 1.35 2009/03/18 16:00:16 cegger 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 #include <sys/conf.h>
49 
50 #include <machine/reg.h>
51 #include <machine/pcb.h>
52 #include <machine/fpu.h>
53 #include <machine/mcontext.h>
54 #include <machine/specialreg.h>
55 #include <machine/vmparam.h>
56 #include <machine/cpufunc.h>
57 
58 /*
59  * To see whether wscons is configured (for virtual console ioctl calls).
60  */
61 #if defined(_KERNEL_OPT)
62 #include "wsdisplay.h"
63 #endif
64 #if (NWSDISPLAY > 0)
65 #include <dev/wscons/wsconsio.h>
66 #include <dev/wscons/wsdisplay_usl_io.h>
67 #endif
68 
69 
70 #include <compat/linux/common/linux_signal.h>
71 #include <compat/linux/common/linux_errno.h>
72 #include <compat/linux/common/linux_exec.h>
73 #include <compat/linux/common/linux_ioctl.h>
74 #include <compat/linux/common/linux_prctl.h>
75 #include <compat/linux/common/linux_machdep.h>
76 #include <compat/linux/common/linux_ipc.h>
77 #include <compat/linux/common/linux_sem.h>
78 #include <compat/linux/linux_syscall.h>
79 #include <compat/linux/linux_syscallargs.h>
80 
81 static void linux_buildcontext(struct lwp *, void *, void *);
82 
83 void
84 linux_setregs(struct lwp *l, struct exec_package *epp, u_long stack)
85 {
86 	struct pcb *pcb = &l->l_addr->u_pcb;
87 	struct trapframe *tf;
88 
89 	/* If we were using the FPU, forget about it. */
90 	if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
91 		fpusave_lwp(l, 0);
92 
93 	l->l_md.md_flags &= ~MDP_USEDFPU;
94 	pcb->pcb_flags = 0;
95 	pcb->pcb_savefpu.fp_fxsave.fx_fcw = __NetBSD_NPXCW__;
96 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr = __INITIAL_MXCSR__;
97 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr_mask = __INITIAL_MXCSR_MASK__;
98 	pcb->pcb_fs = 0;
99 	pcb->pcb_gs = 0;
100 
101 	l->l_proc->p_flag &= ~PK_32;
102 
103 	tf = l->l_md.md_regs;
104 	tf->tf_rax = 0;
105 	tf->tf_rbx = 0;
106 	tf->tf_rcx = epp->ep_entry;
107 	tf->tf_rdx = 0;
108 	tf->tf_rsi = 0;
109 	tf->tf_rdi = 0;
110 	tf->tf_rbp = 0;
111 	tf->tf_rsp = stack;
112 	tf->tf_r8 = 0;
113 	tf->tf_r9 = 0;
114 	tf->tf_r10 = 0;
115 	tf->tf_r11 = 0;
116 	tf->tf_r12 = 0;
117 	tf->tf_r13 = 0;
118 	tf->tf_r14 = 0;
119 	tf->tf_r15 = 0;
120 	tf->tf_rip = epp->ep_entry;
121 	tf->tf_rflags = PSL_USERSET;
122 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
123 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
124 	tf->tf_ds = 0;
125 	tf->tf_es = 0;
126 	tf->tf_fs = 0;
127 	tf->tf_gs = 0;
128 
129 	return;
130 }
131 
132 void
133 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask)
134 {
135 	struct lwp *l = curlwp;
136 	struct proc *p = l->l_proc;
137 	struct sigacts *ps = p->p_sigacts;
138 	int onstack, error;
139 	int sig = ksi->ksi_signo;
140 	struct linux_rt_sigframe *sfp, sigframe;
141 	struct linux__fpstate *fpsp, fpstate;
142 	struct fpreg fpregs;
143 	struct trapframe *tf = l->l_md.md_regs;
144 	sig_t catcher = SIGACTION(p, sig).sa_handler;
145 	linux_sigset_t lmask;
146 	char *sp;
147 
148 	/* Do we need to jump onto the signal stack? */
149 	onstack =
150 	    (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
151 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
152 
153 	/* Allocate space for the signal handler context. */
154 	if (onstack)
155 		sp = ((char *)l->l_sigstk.ss_sp +
156 		    l->l_sigstk.ss_size);
157 	else
158 		sp = (char *)tf->tf_rsp - 128;
159 
160 	/*
161 	 * Save FPU state, if any
162 	 */
163 	if (l->l_md.md_flags & MDP_USEDFPU) {
164 		sp = (char *)
165 		    (((long)sp - sizeof(struct linux__fpstate)) & ~0xfUL);
166 		fpsp = (struct linux__fpstate *)sp;
167 	} else
168 		fpsp = NULL;
169 
170 	/*
171 	 * Populate the rt_sigframe
172 	 */
173 	sp = (char *)
174 	    ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8);
175 	sfp = (struct linux_rt_sigframe *)sp;
176 
177 	memset(&sigframe, 0, sizeof(sigframe));
178 	if (ps->sa_sigdesc[sig].sd_vers != 0)
179 		sigframe.pretcode =
180 		    (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp;
181 	else
182 		sigframe.pretcode = NULL;
183 
184 	/*
185 	 * The user context
186 	 */
187 	sigframe.uc.luc_flags = 0;
188 	sigframe.uc.luc_link = NULL;
189 
190 	/* This is used regardless of SA_ONSTACK in Linux */
191 	sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp;
192 	sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size;
193 	sigframe.uc.luc_stack.ss_flags = 0;
194 	if (l->l_sigstk.ss_flags & SS_ONSTACK)
195 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK;
196 	if (l->l_sigstk.ss_flags & SS_DISABLE)
197 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE;
198 
199 	sigframe.uc.luc_mcontext.r8 = tf->tf_r8;
200 	sigframe.uc.luc_mcontext.r9 = tf->tf_r9;
201 	sigframe.uc.luc_mcontext.r10 = tf->tf_r10;
202 	sigframe.uc.luc_mcontext.r11 = tf->tf_r11;
203 	sigframe.uc.luc_mcontext.r12 = tf->tf_r12;
204 	sigframe.uc.luc_mcontext.r13 = tf->tf_r13;
205 	sigframe.uc.luc_mcontext.r14 = tf->tf_r14;
206 	sigframe.uc.luc_mcontext.r15 = tf->tf_r15;
207 	sigframe.uc.luc_mcontext.rdi = tf->tf_rdi;
208 	sigframe.uc.luc_mcontext.rsi = tf->tf_rsi;
209 	sigframe.uc.luc_mcontext.rbp = tf->tf_rbp;
210 	sigframe.uc.luc_mcontext.rbx = tf->tf_rbx;
211 	sigframe.uc.luc_mcontext.rdx = tf->tf_rdx;
212 	sigframe.uc.luc_mcontext.rax = tf->tf_rax;
213 	sigframe.uc.luc_mcontext.rcx = tf->tf_rcx;
214 	sigframe.uc.luc_mcontext.rsp = tf->tf_rsp;
215 	sigframe.uc.luc_mcontext.rip = tf->tf_rip;
216 	sigframe.uc.luc_mcontext.eflags = tf->tf_rflags;
217 	sigframe.uc.luc_mcontext.cs = tf->tf_cs;
218 	sigframe.uc.luc_mcontext.gs = tf->tf_gs;
219 	sigframe.uc.luc_mcontext.fs = tf->tf_fs;
220 	sigframe.uc.luc_mcontext.err = tf->tf_err;
221 	sigframe.uc.luc_mcontext.trapno = tf->tf_trapno;
222 	native_to_linux_sigset(&lmask, mask);
223 	sigframe.uc.luc_mcontext.oldmask = lmask.sig[0];
224 	sigframe.uc.luc_mcontext.cr2 = (long)l->l_addr->u_pcb.pcb_onfault;
225 	sigframe.uc.luc_mcontext.fpstate = fpsp;
226 	native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask);
227 
228 	/*
229 	 * the siginfo structure
230 	 */
231 	sigframe.info.lsi_signo = native_to_linux_signo[sig];
232 	sigframe.info.lsi_errno = native_to_linux_errno[ksi->ksi_errno];
233 	sigframe.info.lsi_code = native_to_linux_si_code(ksi->ksi_code);
234 
235 	/* XXX This is a rought conversion, taken from i386 code */
236 	switch (sigframe.info.lsi_signo) {
237 	case LINUX_SIGILL:
238 	case LINUX_SIGFPE:
239 	case LINUX_SIGSEGV:
240 	case LINUX_SIGBUS:
241 	case LINUX_SIGTRAP:
242 		sigframe.info._sifields._sigfault._addr = ksi->ksi_addr;
243 		break;
244 	case LINUX_SIGCHLD:
245 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
246 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
247 		sigframe.info._sifields._sigchld._utime = ksi->ksi_utime;
248 		sigframe.info._sifields._sigchld._stime = ksi->ksi_stime;
249 
250 		if (WCOREDUMP(ksi->ksi_status)) {
251 			sigframe.info.lsi_code = LINUX_CLD_DUMPED;
252 			sigframe.info._sifields._sigchld._status =
253 			    _WSTATUS(ksi->ksi_status);
254 		} else if (_WSTATUS(ksi->ksi_status)) {
255 			sigframe.info.lsi_code = LINUX_CLD_KILLED;
256 			sigframe.info._sifields._sigchld._status =
257 			    _WSTATUS(ksi->ksi_status);
258 		} else {
259 			sigframe.info.lsi_code = LINUX_CLD_EXITED;
260 			sigframe.info._sifields._sigchld._status =
261 			    ((ksi->ksi_status & 0xff00U) >> 8);
262 		}
263 		break;
264 	case LINUX_SIGIO:
265 		sigframe.info._sifields._sigpoll._band = ksi->ksi_band;
266 		sigframe.info._sifields._sigpoll._fd = ksi->ksi_fd;
267 		break;
268 	default:
269 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
270 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
271 		if ((sigframe.info.lsi_signo == LINUX_SIGALRM) ||
272 		    (sigframe.info.lsi_signo >= LINUX_SIGRTMIN))
273 			sigframe.info._sifields._timer._sigval.sival_ptr =
274 			     ksi->ksi_value.sival_ptr;
275 		break;
276 	}
277 
278 	sendsig_reset(l, sig);
279 	mutex_exit(p->p_lock);
280 	error = 0;
281 
282 	/*
283 	 * Save FPU state, if any
284 	 */
285 	if (fpsp != NULL) {
286 		(void)process_read_fpregs(l, &fpregs);
287 		memset(&fpstate, 0, sizeof(fpstate));
288 		fpstate.cwd = fpregs.fp_fcw;
289 		fpstate.swd = fpregs.fp_fsw;
290 		fpstate.twd = fpregs.fp_ftw;
291 		fpstate.fop = fpregs.fp_fop;
292 		fpstate.rip = fpregs.fp_rip;
293 		fpstate.rdp = fpregs.fp_rdp;
294 		fpstate.mxcsr = fpregs.fp_mxcsr;
295 		fpstate.mxcsr_mask = fpregs.fp_mxcsr_mask;
296 		memcpy(&fpstate.st_space, &fpregs.fp_st,
297 		    sizeof(fpstate.st_space));
298 		memcpy(&fpstate.xmm_space, &fpregs.fp_xmm,
299 		    sizeof(fpstate.xmm_space));
300 		error = copyout(&fpstate, fpsp, sizeof(fpstate));
301 	}
302 
303 	if (error == 0)
304 		error = copyout(&sigframe, sp, sizeof(sigframe));
305 
306 	mutex_enter(p->p_lock);
307 
308 	if (error != 0) {
309 		sigexit(l, SIGILL);
310 		return;
311 	}
312 
313 	linux_buildcontext(l, catcher, sp);
314 	tf->tf_rdi = sigframe.info.lsi_signo;
315 	tf->tf_rax = 0;
316 	tf->tf_rsi = (long)&sfp->info;
317 	tf->tf_rdx = (long)&sfp->uc;
318 
319 	/*
320 	 * Remember we use signal stack
321 	 */
322 	if (onstack)
323 		l->l_sigstk.ss_flags |= SS_ONSTACK;
324 	return;
325 }
326 
327 int
328 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval)
329 {
330 	printf("linux_sys_modify_ldt\n");
331 	return 0;
332 }
333 
334 int
335 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval)
336 {
337 	return 0;
338 }
339 
340 int
341 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval)
342 {
343 	return 0;
344 }
345 
346 dev_t
347 linux_fakedev(dev_t dev, int raw)
348 {
349 
350        extern const struct cdevsw ptc_cdevsw, pts_cdevsw;
351        const struct cdevsw *cd = cdevsw_lookup(dev);
352 
353        if (raw) {
354 #if (NWSDISPLAY > 0)
355 	       extern const struct cdevsw wsdisplay_cdevsw;
356 	       if (cd == &wsdisplay_cdevsw)
357 		       return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
358 #endif
359        }
360 
361        if (cd == &ptc_cdevsw)
362 	       return makedev(LINUX_PTC_MAJOR, minor(dev));
363        if (cd == &pts_cdevsw)
364 	       return makedev(LINUX_PTS_MAJOR, minor(dev));
365 
366 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
367 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
368 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
369 }
370 
371 int
372 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval)
373 {
374 	return 0;
375 }
376 
377 int
378 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval)
379 {
380 	struct linux_ucontext *luctx;
381 	struct trapframe *tf = l->l_md.md_regs;
382 	struct linux_sigcontext *lsigctx;
383 	struct linux__fpstate fpstate;
384 	struct linux_rt_sigframe frame, *fp;
385 	ucontext_t uctx;
386 	mcontext_t *mctx;
387 	struct fxsave64 *fxarea;
388 	int error;
389 
390 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
391 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
392 		mutex_enter(l->l_proc->p_lock);
393 		sigexit(l, SIGILL);
394 		return error;
395 	}
396 	luctx = &frame.uc;
397 	lsigctx = &luctx->luc_mcontext;
398 
399 	memset(&uctx, 0, sizeof(uctx));
400 	mctx = (mcontext_t *)&uctx.uc_mcontext;
401 	fxarea = (struct fxsave64 *)&mctx->__fpregs;
402 
403 	/*
404 	 * Set the flags. Linux always have CPU, stack and signal state,
405 	 * FPU is optional. uc_flags is not used to tell what we have.
406 	 */
407 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
408 	if (lsigctx->fpstate != NULL)
409 		uctx.uc_flags |= _UC_FPU;
410 	uctx.uc_link = NULL;
411 
412 	/*
413 	 * Signal set
414 	 */
415 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
416 
417 	/*
418 	 * CPU state
419 	 */
420 	mctx->__gregs[_REG_R8] = lsigctx->r8;
421 	mctx->__gregs[_REG_R9] = lsigctx->r9;
422 	mctx->__gregs[_REG_R10] = lsigctx->r10;
423 	mctx->__gregs[_REG_R11] = lsigctx->r11;
424 	mctx->__gregs[_REG_R12] = lsigctx->r12;
425 	mctx->__gregs[_REG_R13] = lsigctx->r13;
426 	mctx->__gregs[_REG_R14] = lsigctx->r14;
427 	mctx->__gregs[_REG_R15] = lsigctx->r15;
428 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
429 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
430 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
431 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
432 	mctx->__gregs[_REG_RAX] = lsigctx->rax;
433 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
434 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
435 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
436 	mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags;
437 	mctx->__gregs[_REG_CS] = lsigctx->cs;
438 	mctx->__gregs[_REG_GS] = lsigctx->gs;
439 	mctx->__gregs[_REG_FS] = lsigctx->fs;
440 	mctx->__gregs[_REG_ERR] = lsigctx->err;
441 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
442 	mctx->__gregs[_REG_ES] = tf->tf_es;
443 	mctx->__gregs[_REG_DS] = tf->tf_ds;
444 	mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */
445 	mctx->__gregs[_REG_SS] = tf->tf_ss;
446 
447 	/*
448 	 * FPU state
449 	 */
450 	if (lsigctx->fpstate != NULL) {
451 		error = copyin(lsigctx->fpstate, &fpstate, sizeof(fpstate));
452 		if (error != 0) {
453 			mutex_enter(l->l_proc->p_lock);
454 			sigexit(l, SIGILL);
455 			return error;
456 		}
457 
458 		fxarea->fx_fcw = fpstate.cwd;
459 		fxarea->fx_fsw = fpstate.swd;
460 		fxarea->fx_ftw = fpstate.twd;
461 		fxarea->fx_fop = fpstate.fop;
462 		fxarea->fx_rip = fpstate.rip;
463 		fxarea->fx_rdp = fpstate.rdp;
464 		fxarea->fx_mxcsr = fpstate.mxcsr;
465 		fxarea->fx_mxcsr_mask = fpstate.mxcsr_mask;
466 		memcpy(&fxarea->fx_st, &fpstate.st_space,
467 		    sizeof(fxarea->fx_st));
468 		memcpy(&fxarea->fx_xmm, &fpstate.xmm_space,
469 		    sizeof(fxarea->fx_xmm));
470 	}
471 
472 	/*
473 	 * And the stack
474 	 */
475 	uctx.uc_stack.ss_flags = 0;
476 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK)
477 		uctx.uc_stack.ss_flags |= SS_ONSTACK;
478 
479 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE)
480 		uctx.uc_stack.ss_flags |= SS_DISABLE;
481 
482 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
483 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
484 
485 	/*
486 	 * And let setucontext deal with that.
487 	 */
488 	mutex_enter(l->l_proc->p_lock);
489 	error = setucontext(l, &uctx);
490 	mutex_exit(l->l_proc->p_lock);
491 	if (error)
492 		return error;
493 
494 	return EJUSTRETURN;
495 }
496 
497 int
498 linux_sys_arch_prctl(struct lwp *l, const struct linux_sys_arch_prctl_args *uap, register_t *retval)
499 {
500 	/* {
501 		syscallarg(int) code;
502 		syscallarg(unsigned long) addr;
503 	} */
504 	struct pcb *pcb = &l->l_addr->u_pcb;
505 	struct trapframe *tf = l->l_md.md_regs;
506 	int error;
507 	uint64_t taddr;
508 
509 	switch(SCARG(uap, code)) {
510 	case LINUX_ARCH_SET_GS:
511 		taddr = SCARG(uap, addr);
512 		if (taddr >= VM_MAXUSER_ADDRESS)
513 			return EINVAL;
514 		pcb->pcb_gs = taddr;
515 		pcb->pcb_flags |= PCB_GS64;
516 		if (l == curlwp)
517 			wrmsr(MSR_KERNELGSBASE, taddr);
518 		break;
519 
520 	case LINUX_ARCH_GET_GS:
521 		if (pcb->pcb_flags & PCB_GS64)
522 			taddr = pcb->pcb_gs;
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 	case LINUX_ARCH_SET_FS:
534 		taddr = SCARG(uap, addr);
535 		if (taddr >= VM_MAXUSER_ADDRESS)
536 			return EINVAL;
537 		pcb->pcb_fs = taddr;
538 		pcb->pcb_flags |= PCB_FS64;
539 		if (l == curlwp)
540 			wrmsr(MSR_FSBASE, taddr);
541 		break;
542 
543 	case LINUX_ARCH_GET_FS:
544 		if (pcb->pcb_flags & PCB_FS64)
545 			taddr = pcb->pcb_fs;
546 		else {
547 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
548 			if (error != 0)
549 				return error;
550 		}
551 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
552 		if (error != 0)
553 			return error;
554 		break;
555 
556 	default:
557 #ifdef DEBUG_LINUX
558 		printf("linux_sys_arch_prctl: unexpected code %d\n",
559 		    SCARG(uap, code));
560 #endif
561 		return EINVAL;
562 	}
563 
564 	return 0;
565 }
566 
567 const int linux_vsyscall_to_syscall[] = {
568 	LINUX_SYS_gettimeofday,
569 	LINUX_SYS_time,
570 	LINUX_SYS_nosys,	/* nosys */
571 	LINUX_SYS_nosys,	/* nosys */
572 };
573 
574 int
575 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
576 {
577 	struct trapframe *tf = arg;
578 	uint64_t retaddr;
579 	int vsyscallnr;
580 
581 	/*
582 	 * Check for a vsyscall. %rip must be the fault address,
583 	 * and the address must be in the Linux vsyscall area.
584 	 * Also, vsyscalls are only done at 1024-byte boundaries.
585 	 */
586 
587 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
588 		return 0;
589 
590 	if (trapaddr != tf->tf_rip)
591 		return 0;
592 
593 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
594 		return 0;
595 
596 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
597 
598 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
599 		return 0;
600 
601 	/*
602 	 * Get the return address from the top of the stack,
603 	 * and fix up the return address.
604 	 * This assumes the faulting instruction was callq *reg,
605 	 * which is the only way that vsyscalls are ever entered.
606 	 */
607 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
608 		return 0;
609 	tf->tf_rip = retaddr;
610 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
611 	tf->tf_rsp += 8;	/* "pop" the return address */
612 
613 #if 0
614 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
615 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
616 	    vsyscallnr, (int)tf->tf_rax);
617 #endif
618 
619 	(*l->l_proc->p_md.md_syscall)(tf);
620 
621 	return 1;
622 }
623 
624 static void
625 linux_buildcontext(struct lwp *l, void *catcher, void *f)
626 {
627 	struct trapframe *tf = l->l_md.md_regs;
628 
629 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
630 	tf->tf_rip = (u_int64_t)catcher;
631 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
632 	tf->tf_rflags &= ~PSL_CLEARSIG;
633 	tf->tf_rsp = (u_int64_t)f;
634 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
635 }
636 
637 void *
638 linux_get_newtls(struct lwp *l)
639 {
640 	struct trapframe *tf = l->l_md.md_regs;
641 
642 	return (void *)tf->tf_r8;
643 }
644 
645 int
646 linux_set_newtls(struct lwp *l, void *tls)
647 {
648 	struct linux_sys_arch_prctl_args cup;
649 	register_t retval;
650 
651 	SCARG(&cup, code) = LINUX_ARCH_SET_FS;
652 	SCARG(&cup, addr) = (unsigned long)tls;
653 
654 	return linux_sys_arch_prctl(l, &cup, &retval);
655 }
656