xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: linux_machdep.c,v 1.26 2007/12/08 18:36:05 dsl 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 1.26 2007/12/08 18:36:05 dsl 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(const ksiginfo_t *ksi, const sigset_t *mask)
136 {
137 	struct lwp *l = curlwp;
138 	struct proc *p = l->l_proc;
139 	struct sigacts *ps = p->p_sigacts;
140 	int onstack, error;
141 	int sig = ksi->ksi_signo;
142 	struct linux_rt_sigframe *sfp, sigframe;
143 	struct linux__fpstate *fpsp, fpstate;
144 	struct fpreg fpregs;
145 	struct trapframe *tf = l->l_md.md_regs;
146 	sig_t catcher = SIGACTION(p, sig).sa_handler;
147 	linux_sigset_t lmask;
148 	char *sp;
149 
150 	/* Do we need to jump onto the signal stack? */
151 	onstack =
152 	    (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
153 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
154 
155 	/* Allocate space for the signal handler context. */
156 	if (onstack)
157 		sp = ((char *)l->l_sigstk.ss_sp +
158 		    l->l_sigstk.ss_size);
159 	else
160 		sp = (char *)tf->tf_rsp - 128;
161 
162 	/*
163 	 * Save FPU state, if any
164 	 */
165 	if (l->l_md.md_flags & MDP_USEDFPU) {
166 		sp = (char *)
167 		    (((long)sp - sizeof(struct linux__fpstate)) & ~0xfUL);
168 		fpsp = (struct linux__fpstate *)sp;
169 	} else
170 		fpsp = NULL;
171 
172 	/*
173 	 * Populate the rt_sigframe
174 	 */
175 	sp = (char *)
176 	    ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8);
177 	sfp = (struct linux_rt_sigframe *)sp;
178 
179 	bzero(&sigframe, sizeof(sigframe));
180 	if (ps->sa_sigdesc[sig].sd_vers != 0)
181 		sigframe.pretcode =
182 		    (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp;
183 	else
184 		sigframe.pretcode = NULL;
185 
186 	/*
187 	 * The user context
188 	 */
189 	sigframe.uc.luc_flags = 0;
190 	sigframe.uc.luc_link = NULL;
191 
192 	/* This is used regardless of SA_ONSTACK in Linux */
193 	sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp;
194 	sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size;
195 	sigframe.uc.luc_stack.ss_flags = 0;
196 	if (l->l_sigstk.ss_flags & SS_ONSTACK)
197 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK;
198 	if (l->l_sigstk.ss_flags & SS_DISABLE)
199 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE;
200 
201 	sigframe.uc.luc_mcontext.r8 = tf->tf_r8;
202 	sigframe.uc.luc_mcontext.r9 = tf->tf_r9;
203 	sigframe.uc.luc_mcontext.r10 = tf->tf_r10;
204 	sigframe.uc.luc_mcontext.r11 = tf->tf_r11;
205 	sigframe.uc.luc_mcontext.r12 = tf->tf_r12;
206 	sigframe.uc.luc_mcontext.r13 = tf->tf_r13;
207 	sigframe.uc.luc_mcontext.r14 = tf->tf_r14;
208 	sigframe.uc.luc_mcontext.r15 = tf->tf_r15;
209 	sigframe.uc.luc_mcontext.rdi = tf->tf_rdi;
210 	sigframe.uc.luc_mcontext.rsi = tf->tf_rsi;
211 	sigframe.uc.luc_mcontext.rbp = tf->tf_rbp;
212 	sigframe.uc.luc_mcontext.rbx = tf->tf_rbx;
213 	sigframe.uc.luc_mcontext.rdx = tf->tf_rdx;
214 	sigframe.uc.luc_mcontext.rax = tf->tf_rax;
215 	sigframe.uc.luc_mcontext.rcx = tf->tf_rcx;
216 	sigframe.uc.luc_mcontext.rsp = tf->tf_rsp;
217 	sigframe.uc.luc_mcontext.rip = tf->tf_rip;
218 	sigframe.uc.luc_mcontext.eflags = tf->tf_rflags;
219 	sigframe.uc.luc_mcontext.cs = tf->tf_cs;
220 	sigframe.uc.luc_mcontext.gs = tf->tf_gs;
221 	sigframe.uc.luc_mcontext.fs = tf->tf_fs;
222 	sigframe.uc.luc_mcontext.err = tf->tf_err;
223 	sigframe.uc.luc_mcontext.trapno = tf->tf_trapno;
224 	native_to_linux_sigset(&lmask, mask);
225 	sigframe.uc.luc_mcontext.oldmask = lmask.sig[0];
226 	sigframe.uc.luc_mcontext.cr2 = (long)l->l_addr->u_pcb.pcb_onfault;
227 	sigframe.uc.luc_mcontext.fpstate = fpsp;
228 	native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask);
229 
230 	/*
231 	 * the siginfo structure
232 	 */
233 	sigframe.info.lsi_signo = native_to_linux_signo[sig];
234 	sigframe.info.lsi_errno = native_to_linux_errno[ksi->ksi_errno];
235 	sigframe.info.lsi_code = ksi->ksi_code;
236 
237 	/* XXX This is a rought conversion, taken from i386 code */
238 	switch (sigframe.info.lsi_signo) {
239 	case LINUX_SIGILL:
240 	case LINUX_SIGFPE:
241 	case LINUX_SIGSEGV:
242 	case LINUX_SIGBUS:
243 	case LINUX_SIGTRAP:
244 		sigframe.info._sifields._sigfault._addr = ksi->ksi_addr;
245 		break;
246 	case LINUX_SIGCHLD:
247 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
248 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
249 		sigframe.info._sifields._sigchld._utime = ksi->ksi_utime;
250 		sigframe.info._sifields._sigchld._stime = ksi->ksi_stime;
251 
252 		if (WCOREDUMP(ksi->ksi_status)) {
253 			sigframe.info.lsi_code = LINUX_CLD_DUMPED;
254 			sigframe.info._sifields._sigchld._status =
255 			    _WSTATUS(ksi->ksi_status);
256 		} else if (_WSTATUS(ksi->ksi_status)) {
257 			sigframe.info.lsi_code = LINUX_CLD_KILLED;
258 			sigframe.info._sifields._sigchld._status =
259 			    _WSTATUS(ksi->ksi_status);
260 		} else {
261 			sigframe.info.lsi_code = LINUX_CLD_EXITED;
262 			sigframe.info._sifields._sigchld._status =
263 			    ((ksi->ksi_status & 0xff00U) >> 8);
264 		}
265 		break;
266 	case LINUX_SIGIO:
267 		sigframe.info._sifields._sigpoll._band = ksi->ksi_band;
268 		sigframe.info._sifields._sigpoll._fd = ksi->ksi_fd;
269 		break;
270 	default:
271 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
272 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
273 		if ((sigframe.info.lsi_signo == LINUX_SIGALRM) ||
274 		    (sigframe.info.lsi_signo >= LINUX_SIGRTMIN))
275 			sigframe.info._sifields._timer._sigval.sival_ptr =
276 			     ksi->ksi_value.sival_ptr;
277 		break;
278 	}
279 
280 	sendsig_reset(l, sig);
281 	mutex_exit(&p->p_smutex);
282 	error = 0;
283 
284 	/*
285 	 * Save FPU state, if any
286 	 */
287 	if (fpsp != NULL) {
288 		(void)process_read_fpregs(l, &fpregs);
289 		bzero(&fpstate, sizeof(fpstate));
290 		fpstate.cwd = fpregs.fp_fcw;
291 		fpstate.swd = fpregs.fp_fsw;
292 		fpstate.twd = fpregs.fp_ftw;
293 		fpstate.fop = fpregs.fp_fop;
294 		fpstate.rip = fpregs.fp_rip;
295 		fpstate.rdp = fpregs.fp_rdp;
296 		fpstate.mxcsr = fpregs.fp_mxcsr;
297 		fpstate.mxcsr_mask = fpregs.fp_mxcsr_mask;
298 		memcpy(&fpstate.st_space, &fpregs.fp_st,
299 		    sizeof(fpstate.st_space));
300 		memcpy(&fpstate.xmm_space, &fpregs.fp_xmm,
301 		    sizeof(fpstate.xmm_space));
302 		error = copyout(&fpstate, fpsp, sizeof(fpstate));
303 	}
304 
305 	if (error == 0)
306 		error = copyout(&sigframe, sp, sizeof(sigframe));
307 
308 	mutex_enter(&p->p_smutex);
309 
310 	if (error != 0) {
311 		sigexit(l, SIGILL);
312 		return;
313 	}
314 
315 	linux_buildcontext(l, catcher, sp);
316 	tf->tf_rdi = sigframe.info.lsi_signo;
317 	tf->tf_rax = 0;
318 	tf->tf_rsi = (long)&sfp->info;
319 	tf->tf_rdx = (long)&sfp->uc;
320 
321 	/*
322 	 * Remember we use signal stack
323 	 */
324 	if (onstack)
325 		l->l_sigstk.ss_flags |= SS_ONSTACK;
326 	return;
327 }
328 
329 int
330 linux_sys_modify_ldt(l, v, retval)
331         struct lwp *l;
332         void *v;
333         register_t *retval;
334 {
335 	printf("linux_sys_modify_ldt\n");
336 	return 0;
337 }
338 
339 int
340 linux_sys_iopl(l, v, retval)
341         struct lwp *l;
342         void *v;
343         register_t *retval;
344 {
345 	return 0;
346 }
347 
348 int
349 linux_sys_ioperm(l, v, retval)
350         struct lwp *l;
351         void *v;
352         register_t *retval;
353 {
354 	return 0;
355 }
356 
357 dev_t
358 linux_fakedev(dev_t dev, int raw)
359 {
360 
361        extern const struct cdevsw ptc_cdevsw, pts_cdevsw;
362        const struct cdevsw *cd = cdevsw_lookup(dev);
363 
364        if (raw) {
365 #if (NWSDISPLAY > 0)
366 	       extern const struct cdevsw wsdisplay_cdevsw;
367 	       if (cd == &wsdisplay_cdevsw)
368 		       return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
369 #endif
370        }
371 
372        if (cd == &ptc_cdevsw)
373 	       return makedev(LINUX_PTC_MAJOR, minor(dev));
374        if (cd == &pts_cdevsw)
375 	       return makedev(LINUX_PTS_MAJOR, minor(dev));
376 
377 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
378 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
379 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
380 }
381 
382 int
383 linux_machdepioctl(l, v, retval)
384         struct lwp *l;
385         void *v;
386         register_t *retval;
387 {
388 	return 0;
389 }
390 
391 int
392 linux_sys_rt_sigreturn(l, v, retval)
393         struct lwp *l;
394         void *v;
395         register_t *retval;
396 {
397 	struct linux_ucontext *luctx;
398 	struct trapframe *tf = l->l_md.md_regs;
399 	struct linux_sigcontext *lsigctx;
400 	struct linux__fpstate fpstate;
401 	struct linux_rt_sigframe frame, *fp;
402 	ucontext_t uctx;
403 	mcontext_t *mctx;
404 	struct fxsave64 *fxarea;
405 	int error;
406 
407 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
408 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
409 		mutex_enter(&l->l_proc->p_smutex);
410 		sigexit(l, SIGILL);
411 		return error;
412 	}
413 	luctx = &frame.uc;
414 	lsigctx = &luctx->luc_mcontext;
415 
416 	bzero(&uctx, sizeof(uctx));
417 	mctx = (mcontext_t *)&uctx.uc_mcontext;
418 	fxarea = (struct fxsave64 *)&mctx->__fpregs;
419 
420 	/*
421 	 * Set the flags. Linux always have CPU, stack and signal state,
422 	 * FPU is optional. uc_flags is not used to tell what we have.
423 	 */
424 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
425 	if (lsigctx->fpstate != NULL)
426 		uctx.uc_flags |= _UC_FPU;
427 	uctx.uc_link = NULL;
428 
429 	/*
430 	 * Signal set
431 	 */
432 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
433 
434 	/*
435 	 * CPU state
436 	 */
437 	mctx->__gregs[_REG_R8] = lsigctx->r8;
438 	mctx->__gregs[_REG_R9] = lsigctx->r9;
439 	mctx->__gregs[_REG_R10] = lsigctx->r10;
440 	mctx->__gregs[_REG_R11] = lsigctx->r11;
441 	mctx->__gregs[_REG_R12] = lsigctx->r12;
442 	mctx->__gregs[_REG_R13] = lsigctx->r13;
443 	mctx->__gregs[_REG_R14] = lsigctx->r14;
444 	mctx->__gregs[_REG_R15] = lsigctx->r15;
445 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
446 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
447 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
448 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
449 	mctx->__gregs[_REG_RAX] = lsigctx->rax;
450 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
451 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
452 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
453 	mctx->__gregs[_REG_RFL] = lsigctx->eflags;
454 	mctx->__gregs[_REG_CS] = lsigctx->cs;
455 	mctx->__gregs[_REG_GS] = lsigctx->gs;
456 	mctx->__gregs[_REG_FS] = lsigctx->fs;
457 	mctx->__gregs[_REG_ERR] = lsigctx->err;
458 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
459 	mctx->__gregs[_REG_ES] = tf->tf_es;
460 	mctx->__gregs[_REG_DS] = tf->tf_ds;
461 	mctx->__gregs[_REG_URSP] = lsigctx->rsp; /* XXX */
462 	mctx->__gregs[_REG_SS] = tf->tf_ss;
463 
464 	/*
465 	 * FPU state
466 	 */
467 	if (lsigctx->fpstate != NULL) {
468 		error = copyin(lsigctx->fpstate, &fpstate, sizeof(fpstate));
469 		if (error != 0) {
470 			mutex_enter(&l->l_proc->p_smutex);
471 			sigexit(l, SIGILL);
472 			return error;
473 		}
474 
475 		fxarea->fx_fcw = fpstate.cwd;
476 		fxarea->fx_fsw = fpstate.swd;
477 		fxarea->fx_ftw = fpstate.twd;
478 		fxarea->fx_fop = fpstate.fop;
479 		fxarea->fx_rip = fpstate.rip;
480 		fxarea->fx_rdp = fpstate.rdp;
481 		fxarea->fx_mxcsr = fpstate.mxcsr;
482 		fxarea->fx_mxcsr_mask = fpstate.mxcsr_mask;
483 		memcpy(&fxarea->fx_st, &fpstate.st_space,
484 		    sizeof(fxarea->fx_st));
485 		memcpy(&fxarea->fx_xmm, &fpstate.xmm_space,
486 		    sizeof(fxarea->fx_xmm));
487 	}
488 
489 	/*
490 	 * And the stack
491 	 */
492 	uctx.uc_stack.ss_flags = 0;
493 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK);
494 		uctx.uc_stack.ss_flags = SS_ONSTACK;
495 
496 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE);
497 		uctx.uc_stack.ss_flags = SS_DISABLE;
498 
499 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
500 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
501 
502 	/*
503 	 * And let setucontext deal with that.
504 	 */
505 	mutex_enter(&l->l_proc->p_smutex);
506 	error = setucontext(l, &uctx);
507 	mutex_exit(&l->l_proc->p_smutex);
508 	if (error)
509 		return error;
510 
511 	return EJUSTRETURN;
512 }
513 
514 int
515 linux_sys_arch_prctl(struct lwp *l, void *v, register_t *retval)
516 {
517 	struct linux_sys_arch_prctl_args /* {
518 		syscallarg(int) code;
519 		syscallarg(unsigned long) addr;
520 	} */ *uap = v;
521 	struct pcb *pcb = &l->l_addr->u_pcb;
522 	struct trapframe *tf = l->l_md.md_regs;
523 	int error;
524 	uint64_t taddr;
525 
526 	switch(SCARG(uap, code)) {
527 	case LINUX_ARCH_SET_GS:
528 		taddr = SCARG(uap, addr);
529 		if (taddr >= VM_MAXUSER_ADDRESS)
530 			return EINVAL;
531 		pcb->pcb_gs = taddr;
532 		pcb->pcb_flags |= PCB_GS64;
533 		if (l == curlwp)
534 			wrmsr(MSR_KERNELGSBASE, taddr);
535 		break;
536 
537 	case LINUX_ARCH_GET_GS:
538 		if (pcb->pcb_flags & PCB_GS64)
539 			taddr = pcb->pcb_gs;
540 		else {
541 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
542 			if (error != 0)
543 				return error;
544 		}
545 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
546 		if (error != 0)
547 			return error;
548 		break;
549 
550 	case LINUX_ARCH_SET_FS:
551 		taddr = SCARG(uap, addr);
552 		if (taddr >= VM_MAXUSER_ADDRESS)
553 			return EINVAL;
554 		pcb->pcb_fs = taddr;
555 		pcb->pcb_flags |= PCB_FS64;
556 		if (l == curlwp)
557 			wrmsr(MSR_FSBASE, taddr);
558 		break;
559 
560 	case LINUX_ARCH_GET_FS:
561 		if (pcb->pcb_flags & PCB_FS64)
562 			taddr = pcb->pcb_fs;
563 		else {
564 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
565 			if (error != 0)
566 				return error;
567 		}
568 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
569 		if (error != 0)
570 			return error;
571 		break;
572 
573 	default:
574 #ifdef DEBUG_LINUX
575 		printf("linux_sys_arch_prctl: unexpected code %d\n",
576 		    SCARG(uap, code));
577 #endif
578 		return EINVAL;
579 	}
580 
581 	return 0;
582 }
583 
584 const int linux_vsyscall_to_syscall[] = {
585 	LINUX_SYS_gettimeofday,
586 	LINUX_SYS_time,
587 	LINUX_SYS_nosys,
588 	LINUX_SYS_nosys,
589 };
590 
591 int
592 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
593 {
594 	struct trapframe *tf = arg;
595 	uint64_t retaddr;
596 	int vsyscallnr;
597 
598 	/*
599 	 * Check for a vsyscall. %rip must be the fault address,
600 	 * and the address must be in the Linux vsyscall area.
601 	 * Also, vsyscalls are only done at 1024-byte boundaries.
602 	 */
603 
604 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
605 		return 0;
606 
607 	if (trapaddr != tf->tf_rip)
608 		return 0;
609 
610 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
611 		return 0;
612 
613 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
614 
615 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
616 		return 0;
617 
618 	/*
619 	 * Get the return address from the top of the stack,
620 	 * and fix up the return address.
621 	 * This assumes the faulting instruction was callq *reg,
622 	 * which is the only way that vsyscalls are ever entered.
623 	 */
624 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
625 		return 0;
626 	tf->tf_rip = retaddr;
627 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
628 	tf->tf_rsp += 8;	/* "pop" the return address */
629 
630 #if 0
631 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
632 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
633 	    vsyscallnr, (int)tf->tf_rax);
634 #endif
635 
636 	(*l->l_proc->p_md.md_syscall)(tf);
637 
638 	return 1;
639 }
640 
641 static void
642 linux_buildcontext(struct lwp *l, void *catcher, void *f)
643 {
644 	struct trapframe *tf = l->l_md.md_regs;
645 
646 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
647 	tf->tf_rip = (u_int64_t)catcher;
648 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
649 	tf->tf_rflags &= ~(PSL_T|PSL_VM|PSL_AC);
650 	tf->tf_rsp = (u_int64_t)f;
651 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
652 }
653 
654 void *
655 linux_get_newtls(struct lwp *l)
656 {
657 	struct trapframe *tf = l->l_md.md_regs;
658 
659 	return (void *)tf->tf_r8;
660 }
661 
662 int
663 linux_set_newtls(struct lwp *l, void *tls)
664 {
665 	struct linux_sys_arch_prctl_args cup;
666 	register_t retval;
667 
668 	SCARG(&cup, code) = LINUX_ARCH_SET_FS;
669 	SCARG(&cup, addr) = (unsigned long)tls;
670 
671 	return linux_sys_arch_prctl(l, &cup, &retval);
672 }
673