xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision 0df165c04d0a9ca1adde9ed2b890344c937954a6)
1 /*	$NetBSD: linux_machdep.c,v 1.25 2007/10/19 18:52:10 njoly 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 1.25 2007/10/19 18:52:10 njoly 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 
57 /*
58  * To see whether wscons is configured (for virtual console ioctl calls).
59  */
60 #if defined(_KERNEL_OPT)
61 #include "wsdisplay.h"
62 #endif
63 #if (NWSDISPLAY > 0)
64 #include <dev/wscons/wsconsio.h>
65 #include <dev/wscons/wsdisplay_usl_io.h>
66 #endif
67 
68 
69 #include <compat/linux/common/linux_signal.h>
70 #include <compat/linux/common/linux_errno.h>
71 #include <compat/linux/common/linux_exec.h>
72 #include <compat/linux/common/linux_ioctl.h>
73 #include <compat/linux/common/linux_prctl.h>
74 #include <compat/linux/common/linux_machdep.h>
75 #include <compat/linux/common/linux_ipc.h>
76 #include <compat/linux/common/linux_sem.h>
77 #include <compat/linux/linux_syscall.h>
78 #include <compat/linux/linux_syscallargs.h>
79 
80 static void linux_buildcontext(struct lwp *, void *, void *);
81 
82 void
83 linux_setregs(l, epp, stack)
84         struct lwp *l;
85 	struct exec_package *epp;
86 	u_long stack;
87 {
88 	struct pcb *pcb = &l->l_addr->u_pcb;
89 	struct trapframe *tf;
90 
91 	/* If we were using the FPU, forget about it. */
92 	if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
93 		fpusave_lwp(l, 0);
94 
95 	l->l_md.md_flags &= ~MDP_USEDFPU;
96 	pcb->pcb_flags = 0;
97 	pcb->pcb_savefpu.fp_fxsave.fx_fcw = __NetBSD_NPXCW__;
98 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr = __INITIAL_MXCSR__;
99 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr_mask = __INITIAL_MXCSR_MASK__;
100 	pcb->pcb_fs = 0;
101 	pcb->pcb_gs = 0;
102 
103 	l->l_proc->p_flag &= ~PK_32;
104 
105 	tf = l->l_md.md_regs;
106 	tf->tf_rax = 0;
107 	tf->tf_rbx = 0;
108 	tf->tf_rcx = epp->ep_entry;
109 	tf->tf_rdx = 0;
110 	tf->tf_rsi = 0;
111 	tf->tf_rdi = 0;
112 	tf->tf_rbp = 0;
113 	tf->tf_rsp = stack;
114 	tf->tf_r8 = 0;
115 	tf->tf_r9 = 0;
116 	tf->tf_r10 = 0;
117 	tf->tf_r11 = 0;
118 	tf->tf_r12 = 0;
119 	tf->tf_r13 = 0;
120 	tf->tf_r14 = 0;
121 	tf->tf_r15 = 0;
122 	tf->tf_rip = epp->ep_entry;
123 	tf->tf_rflags = PSL_USERSET;
124 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
125 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
126 	tf->tf_ds = 0;
127 	tf->tf_es = 0;
128 	tf->tf_fs = 0;
129 	tf->tf_gs = 0;
130 
131 	return;
132 }
133 
134 void
135 linux_sendsig(ksi, mask)
136 	const ksiginfo_t *ksi;
137 	const sigset_t *mask;
138 {
139 	struct lwp *l = curlwp;
140 	struct proc *p = l->l_proc;
141 	struct sigacts *ps = p->p_sigacts;
142 	int onstack, error;
143 	int sig = ksi->ksi_signo;
144 	struct linux_rt_sigframe *sfp, sigframe;
145 	struct linux__fpstate *fpsp, fpstate;
146 	struct fpreg fpregs;
147 	struct trapframe *tf = l->l_md.md_regs;
148 	sig_t catcher = SIGACTION(p, sig).sa_handler;
149 	linux_sigset_t lmask;
150 	char *sp;
151 
152 	/* Do we need to jump onto the signal stack? */
153 	onstack =
154 	    (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
155 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
156 
157 	/* Allocate space for the signal handler context. */
158 	if (onstack)
159 		sp = ((char *)l->l_sigstk.ss_sp +
160 		    l->l_sigstk.ss_size);
161 	else
162 		sp = (char *)tf->tf_rsp - 128;
163 
164 	/*
165 	 * Save FPU state, if any
166 	 */
167 	if (l->l_md.md_flags & MDP_USEDFPU) {
168 		sp = (char *)
169 		    (((long)sp - sizeof(struct linux__fpstate)) & ~0xfUL);
170 		fpsp = (struct linux__fpstate *)sp;
171 	} else
172 		fpsp = NULL;
173 
174 	/*
175 	 * Populate the rt_sigframe
176 	 */
177 	sp = (char *)
178 	    ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8);
179 	sfp = (struct linux_rt_sigframe *)sp;
180 
181 	bzero(&sigframe, sizeof(sigframe));
182 	if (ps->sa_sigdesc[sig].sd_vers != 0)
183 		sigframe.pretcode =
184 		    (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp;
185 	else
186 		sigframe.pretcode = NULL;
187 
188 	/*
189 	 * The user context
190 	 */
191 	sigframe.uc.luc_flags = 0;
192 	sigframe.uc.luc_link = NULL;
193 
194 	/* This is used regardless of SA_ONSTACK in Linux */
195 	sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp;
196 	sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size;
197 	sigframe.uc.luc_stack.ss_flags = 0;
198 	if (l->l_sigstk.ss_flags & SS_ONSTACK)
199 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK;
200 	if (l->l_sigstk.ss_flags & SS_DISABLE)
201 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE;
202 
203 	sigframe.uc.luc_mcontext.r8 = tf->tf_r8;
204 	sigframe.uc.luc_mcontext.r9 = tf->tf_r9;
205 	sigframe.uc.luc_mcontext.r10 = tf->tf_r10;
206 	sigframe.uc.luc_mcontext.r11 = tf->tf_r11;
207 	sigframe.uc.luc_mcontext.r12 = tf->tf_r12;
208 	sigframe.uc.luc_mcontext.r13 = tf->tf_r13;
209 	sigframe.uc.luc_mcontext.r14 = tf->tf_r14;
210 	sigframe.uc.luc_mcontext.r15 = tf->tf_r15;
211 	sigframe.uc.luc_mcontext.rdi = tf->tf_rdi;
212 	sigframe.uc.luc_mcontext.rsi = tf->tf_rsi;
213 	sigframe.uc.luc_mcontext.rbp = tf->tf_rbp;
214 	sigframe.uc.luc_mcontext.rbx = tf->tf_rbx;
215 	sigframe.uc.luc_mcontext.rdx = tf->tf_rdx;
216 	sigframe.uc.luc_mcontext.rax = tf->tf_rax;
217 	sigframe.uc.luc_mcontext.rcx = tf->tf_rcx;
218 	sigframe.uc.luc_mcontext.rsp = tf->tf_rsp;
219 	sigframe.uc.luc_mcontext.rip = tf->tf_rip;
220 	sigframe.uc.luc_mcontext.eflags = tf->tf_rflags;
221 	sigframe.uc.luc_mcontext.cs = tf->tf_cs;
222 	sigframe.uc.luc_mcontext.gs = tf->tf_gs;
223 	sigframe.uc.luc_mcontext.fs = tf->tf_fs;
224 	sigframe.uc.luc_mcontext.err = tf->tf_err;
225 	sigframe.uc.luc_mcontext.trapno = tf->tf_trapno;
226 	native_to_linux_sigset(&lmask, mask);
227 	sigframe.uc.luc_mcontext.oldmask = lmask.sig[0];
228 	sigframe.uc.luc_mcontext.cr2 = (long)l->l_addr->u_pcb.pcb_onfault;
229 	sigframe.uc.luc_mcontext.fpstate = fpsp;
230 	native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask);
231 
232 	/*
233 	 * the siginfo structure
234 	 */
235 	sigframe.info.lsi_signo = native_to_linux_signo[sig];
236 	sigframe.info.lsi_errno = native_to_linux_errno[ksi->ksi_errno];
237 	sigframe.info.lsi_code = ksi->ksi_code;
238 
239 	/* XXX This is a rought conversion, taken from i386 code */
240 	switch (sigframe.info.lsi_signo) {
241 	case LINUX_SIGILL:
242 	case LINUX_SIGFPE:
243 	case LINUX_SIGSEGV:
244 	case LINUX_SIGBUS:
245 	case LINUX_SIGTRAP:
246 		sigframe.info._sifields._sigfault._addr = ksi->ksi_addr;
247 		break;
248 	case LINUX_SIGCHLD:
249 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
250 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
251 		sigframe.info._sifields._sigchld._utime = ksi->ksi_utime;
252 		sigframe.info._sifields._sigchld._stime = ksi->ksi_stime;
253 
254 		if (WCOREDUMP(ksi->ksi_status)) {
255 			sigframe.info.lsi_code = LINUX_CLD_DUMPED;
256 			sigframe.info._sifields._sigchld._status =
257 			    _WSTATUS(ksi->ksi_status);
258 		} else if (_WSTATUS(ksi->ksi_status)) {
259 			sigframe.info.lsi_code = LINUX_CLD_KILLED;
260 			sigframe.info._sifields._sigchld._status =
261 			    _WSTATUS(ksi->ksi_status);
262 		} else {
263 			sigframe.info.lsi_code = LINUX_CLD_EXITED;
264 			sigframe.info._sifields._sigchld._status =
265 			    ((ksi->ksi_status & 0xff00U) >> 8);
266 		}
267 		break;
268 	case LINUX_SIGIO:
269 		sigframe.info._sifields._sigpoll._band = ksi->ksi_band;
270 		sigframe.info._sifields._sigpoll._fd = ksi->ksi_fd;
271 		break;
272 	default:
273 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
274 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
275 		if ((sigframe.info.lsi_signo == LINUX_SIGALRM) ||
276 		    (sigframe.info.lsi_signo >= LINUX_SIGRTMIN))
277 			sigframe.info._sifields._timer._sigval.sival_ptr =
278 			     ksi->ksi_value.sival_ptr;
279 		break;
280 	}
281 
282 	sendsig_reset(l, sig);
283 	mutex_exit(&p->p_smutex);
284 	error = 0;
285 
286 	/*
287 	 * Save FPU state, if any
288 	 */
289 	if (fpsp != NULL) {
290 		(void)process_read_fpregs(l, &fpregs);
291 		bzero(&fpstate, sizeof(fpstate));
292 		fpstate.cwd = fpregs.fp_fcw;
293 		fpstate.swd = fpregs.fp_fsw;
294 		fpstate.twd = fpregs.fp_ftw;
295 		fpstate.fop = fpregs.fp_fop;
296 		fpstate.rip = fpregs.fp_rip;
297 		fpstate.rdp = fpregs.fp_rdp;
298 		fpstate.mxcsr = fpregs.fp_mxcsr;
299 		fpstate.mxcsr_mask = fpregs.fp_mxcsr_mask;
300 		memcpy(&fpstate.st_space, &fpregs.fp_st,
301 		    sizeof(fpstate.st_space));
302 		memcpy(&fpstate.xmm_space, &fpregs.fp_xmm,
303 		    sizeof(fpstate.xmm_space));
304 		error = copyout(&fpstate, fpsp, sizeof(fpstate));
305 	}
306 
307 	if (error == 0)
308 		error = copyout(&sigframe, sp, sizeof(sigframe));
309 
310 	mutex_enter(&p->p_smutex);
311 
312 	if (error != 0) {
313 		sigexit(l, SIGILL);
314 		return;
315 	}
316 
317 	linux_buildcontext(l, catcher, sp);
318 	tf->tf_rdi = sigframe.info.lsi_signo;
319 	tf->tf_rax = 0;
320 	tf->tf_rsi = (long)&sfp->info;
321 	tf->tf_rdx = (long)&sfp->uc;
322 
323 	/*
324 	 * Remember we use signal stack
325 	 */
326 	if (onstack)
327 		l->l_sigstk.ss_flags |= SS_ONSTACK;
328 	return;
329 }
330 
331 int
332 linux_sys_modify_ldt(l, v, retval)
333         struct lwp *l;
334         void *v;
335         register_t *retval;
336 {
337 	printf("linux_sys_modify_ldt\n");
338 	return 0;
339 }
340 
341 int
342 linux_sys_iopl(l, v, retval)
343         struct lwp *l;
344         void *v;
345         register_t *retval;
346 {
347 	return 0;
348 }
349 
350 int
351 linux_sys_ioperm(l, v, retval)
352         struct lwp *l;
353         void *v;
354         register_t *retval;
355 {
356 	return 0;
357 }
358 
359 dev_t
360 linux_fakedev(dev, raw)
361         dev_t dev;
362 	int raw;
363 {
364 
365        extern const struct cdevsw ptc_cdevsw, pts_cdevsw;
366        const struct cdevsw *cd = cdevsw_lookup(dev);
367 
368        if (raw) {
369 #if (NWSDISPLAY > 0)
370 	       extern const struct cdevsw wsdisplay_cdevsw;
371 	       if (cd == &wsdisplay_cdevsw)
372 		       return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
373 #endif
374        }
375 
376        if (cd == &ptc_cdevsw)
377 	       return makedev(LINUX_PTC_MAJOR, minor(dev));
378        if (cd == &pts_cdevsw)
379 	       return makedev(LINUX_PTS_MAJOR, minor(dev));
380 
381 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
382 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
383 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
384 }
385 
386 int
387 linux_machdepioctl(l, v, retval)
388         struct lwp *l;
389         void *v;
390         register_t *retval;
391 {
392 	return 0;
393 }
394 
395 int
396 linux_sys_rt_sigreturn(l, v, retval)
397         struct lwp *l;
398         void *v;
399         register_t *retval;
400 {
401 	struct linux_ucontext *luctx;
402 	struct trapframe *tf = l->l_md.md_regs;
403 	struct linux_sigcontext *lsigctx;
404 	struct linux__fpstate fpstate;
405 	struct linux_rt_sigframe frame, *fp;
406 	ucontext_t uctx;
407 	mcontext_t *mctx;
408 	struct fxsave64 *fxarea;
409 	int error;
410 
411 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
412 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
413 		mutex_enter(&l->l_proc->p_smutex);
414 		sigexit(l, SIGILL);
415 		return error;
416 	}
417 	luctx = &frame.uc;
418 	lsigctx = &luctx->luc_mcontext;
419 
420 	bzero(&uctx, sizeof(uctx));
421 	mctx = (mcontext_t *)&uctx.uc_mcontext;
422 	fxarea = (struct fxsave64 *)&mctx->__fpregs;
423 
424 	/*
425 	 * Set the flags. Linux always have CPU, stack and signal state,
426 	 * FPU is optional. uc_flags is not used to tell what we have.
427 	 */
428 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
429 	if (lsigctx->fpstate != NULL)
430 		uctx.uc_flags |= _UC_FPU;
431 	uctx.uc_link = NULL;
432 
433 	/*
434 	 * Signal set
435 	 */
436 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
437 
438 	/*
439 	 * CPU state
440 	 */
441 	mctx->__gregs[_REG_R8] = lsigctx->r8;
442 	mctx->__gregs[_REG_R9] = lsigctx->r9;
443 	mctx->__gregs[_REG_R10] = lsigctx->r10;
444 	mctx->__gregs[_REG_R11] = lsigctx->r11;
445 	mctx->__gregs[_REG_R12] = lsigctx->r12;
446 	mctx->__gregs[_REG_R13] = lsigctx->r13;
447 	mctx->__gregs[_REG_R14] = lsigctx->r14;
448 	mctx->__gregs[_REG_R15] = lsigctx->r15;
449 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
450 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
451 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
452 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
453 	mctx->__gregs[_REG_RAX] = lsigctx->rax;
454 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
455 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
456 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
457 	mctx->__gregs[_REG_RFL] = lsigctx->eflags;
458 	mctx->__gregs[_REG_CS] = lsigctx->cs;
459 	mctx->__gregs[_REG_GS] = lsigctx->gs;
460 	mctx->__gregs[_REG_FS] = lsigctx->fs;
461 	mctx->__gregs[_REG_ERR] = lsigctx->err;
462 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
463 	mctx->__gregs[_REG_ES] = tf->tf_es;
464 	mctx->__gregs[_REG_DS] = tf->tf_ds;
465 	mctx->__gregs[_REG_URSP] = lsigctx->rsp; /* XXX */
466 	mctx->__gregs[_REG_SS] = tf->tf_ss;
467 
468 	/*
469 	 * FPU state
470 	 */
471 	if (lsigctx->fpstate != NULL) {
472 		error = copyin(lsigctx->fpstate, &fpstate, sizeof(fpstate));
473 		if (error != 0) {
474 			mutex_enter(&l->l_proc->p_smutex);
475 			sigexit(l, SIGILL);
476 			return error;
477 		}
478 
479 		fxarea->fx_fcw = fpstate.cwd;
480 		fxarea->fx_fsw = fpstate.swd;
481 		fxarea->fx_ftw = fpstate.twd;
482 		fxarea->fx_fop = fpstate.fop;
483 		fxarea->fx_rip = fpstate.rip;
484 		fxarea->fx_rdp = fpstate.rdp;
485 		fxarea->fx_mxcsr = fpstate.mxcsr;
486 		fxarea->fx_mxcsr_mask = fpstate.mxcsr_mask;
487 		memcpy(&fxarea->fx_st, &fpstate.st_space,
488 		    sizeof(fxarea->fx_st));
489 		memcpy(&fxarea->fx_xmm, &fpstate.xmm_space,
490 		    sizeof(fxarea->fx_xmm));
491 	}
492 
493 	/*
494 	 * And the stack
495 	 */
496 	uctx.uc_stack.ss_flags = 0;
497 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK);
498 		uctx.uc_stack.ss_flags = SS_ONSTACK;
499 
500 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE);
501 		uctx.uc_stack.ss_flags = SS_DISABLE;
502 
503 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
504 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
505 
506 	/*
507 	 * And let setucontext deal with that.
508 	 */
509 	mutex_enter(&l->l_proc->p_smutex);
510 	error = setucontext(l, &uctx);
511 	mutex_exit(&l->l_proc->p_smutex);
512 	if (error)
513 		return error;
514 
515 	return EJUSTRETURN;
516 }
517 
518 int
519 linux_sys_arch_prctl(l, v, retval)
520 	struct lwp *l;
521 	void *v;
522 	register_t *retval;
523 {
524 	struct linux_sys_arch_prctl_args /* {
525 		syscallarg(int) code;
526 		syscallarg(unsigned long) addr;
527 	} */ *uap = v;
528 	struct pcb *pcb = &l->l_addr->u_pcb;
529 	struct trapframe *tf = l->l_md.md_regs;
530 	int error;
531 	uint64_t taddr;
532 
533 	switch(SCARG(uap, code)) {
534 	case LINUX_ARCH_SET_GS:
535 		taddr = SCARG(uap, addr);
536 		if (taddr >= VM_MAXUSER_ADDRESS)
537 			return EINVAL;
538 		pcb->pcb_gs = taddr;
539 		pcb->pcb_flags |= PCB_GS64;
540 		if (l == curlwp)
541 			wrmsr(MSR_KERNELGSBASE, taddr);
542 		break;
543 
544 	case LINUX_ARCH_GET_GS:
545 		if (pcb->pcb_flags & PCB_GS64)
546 			taddr = pcb->pcb_gs;
547 		else {
548 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
549 			if (error != 0)
550 				return error;
551 		}
552 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
553 		if (error != 0)
554 			return error;
555 		break;
556 
557 	case LINUX_ARCH_SET_FS:
558 		taddr = SCARG(uap, addr);
559 		if (taddr >= VM_MAXUSER_ADDRESS)
560 			return EINVAL;
561 		pcb->pcb_fs = taddr;
562 		pcb->pcb_flags |= PCB_FS64;
563 		if (l == curlwp)
564 			wrmsr(MSR_FSBASE, taddr);
565 		break;
566 
567 	case LINUX_ARCH_GET_FS:
568 		if (pcb->pcb_flags & PCB_FS64)
569 			taddr = pcb->pcb_fs;
570 		else {
571 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
572 			if (error != 0)
573 				return error;
574 		}
575 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
576 		if (error != 0)
577 			return error;
578 		break;
579 
580 	default:
581 #ifdef DEBUG_LINUX
582 		printf("linux_sys_arch_prctl: unexpected code %d\n",
583 		    SCARG(uap, code));
584 #endif
585 		return EINVAL;
586 	}
587 
588 	return 0;
589 }
590 
591 const int linux_vsyscall_to_syscall[] = {
592 	LINUX_SYS_gettimeofday,
593 	LINUX_SYS_time,
594 	LINUX_SYS_nosys,
595 	LINUX_SYS_nosys,
596 };
597 
598 int
599 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
600 {
601 	struct trapframe *tf = arg;
602 	uint64_t retaddr;
603 	int vsyscallnr;
604 
605 	/*
606 	 * Check for a vsyscall. %rip must be the fault address,
607 	 * and the address must be in the Linux vsyscall area.
608 	 * Also, vsyscalls are only done at 1024-byte boundaries.
609 	 */
610 
611 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
612 		return 0;
613 
614 	if (trapaddr != tf->tf_rip)
615 		return 0;
616 
617 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
618 		return 0;
619 
620 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
621 
622 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
623 		return 0;
624 
625 	/*
626 	 * Get the return address from the top of the stack,
627 	 * and fix up the return address.
628 	 * This assumes the faulting instruction was callq *reg,
629 	 * which is the only way that vsyscalls are ever entered.
630 	 */
631 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
632 		return 0;
633 	tf->tf_rip = retaddr;
634 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
635 	tf->tf_rsp += 8;	/* "pop" the return address */
636 
637 #if 0
638 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
639 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
640 	    vsyscallnr, (int)tf->tf_rax);
641 #endif
642 
643 	(*l->l_proc->p_md.md_syscall)(tf);
644 
645 	return 1;
646 }
647 
648 static void
649 linux_buildcontext(struct lwp *l, void *catcher, void *f)
650 {
651 	struct trapframe *tf = l->l_md.md_regs;
652 
653 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
654 	tf->tf_rip = (u_int64_t)catcher;
655 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
656 	tf->tf_rflags &= ~(PSL_T|PSL_VM|PSL_AC);
657 	tf->tf_rsp = (u_int64_t)f;
658 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
659 }
660 
661 void *
662 linux_get_newtls(l)
663 	struct lwp *l;
664 {
665 	struct trapframe *tf = l->l_md.md_regs;
666 
667 	return (void *)tf->tf_r8;
668 }
669 
670 int
671 linux_set_newtls(l, tls)
672 	struct lwp *l;
673 	void *tls;
674 {
675 	struct linux_sys_arch_prctl_args cup;
676 	register_t retval;
677 
678 	SCARG(&cup, code) = LINUX_ARCH_SET_FS;
679 	SCARG(&cup, addr) = (unsigned long)tls;
680 
681 	return linux_sys_arch_prctl(l, &cup, &retval);
682 }
683