xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: linux_machdep.c,v 1.28 2008/01/05 19:11:53 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 1.28 2008/01/05 19:11:53 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(struct lwp *l, struct exec_package *epp, u_long stack)
84 {
85 	struct pcb *pcb = &l->l_addr->u_pcb;
86 	struct trapframe *tf;
87 
88 	/* If we were using the FPU, forget about it. */
89 	if (l->l_addr->u_pcb.pcb_fpcpu != NULL)
90 		fpusave_lwp(l, 0);
91 
92 	l->l_md.md_flags &= ~MDP_USEDFPU;
93 	pcb->pcb_flags = 0;
94 	pcb->pcb_savefpu.fp_fxsave.fx_fcw = __NetBSD_NPXCW__;
95 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr = __INITIAL_MXCSR__;
96 	pcb->pcb_savefpu.fp_fxsave.fx_mxcsr_mask = __INITIAL_MXCSR_MASK__;
97 	pcb->pcb_fs = 0;
98 	pcb->pcb_gs = 0;
99 
100 	l->l_proc->p_flag &= ~PK_32;
101 
102 	tf = l->l_md.md_regs;
103 	tf->tf_rax = 0;
104 	tf->tf_rbx = 0;
105 	tf->tf_rcx = epp->ep_entry;
106 	tf->tf_rdx = 0;
107 	tf->tf_rsi = 0;
108 	tf->tf_rdi = 0;
109 	tf->tf_rbp = 0;
110 	tf->tf_rsp = stack;
111 	tf->tf_r8 = 0;
112 	tf->tf_r9 = 0;
113 	tf->tf_r10 = 0;
114 	tf->tf_r11 = 0;
115 	tf->tf_r12 = 0;
116 	tf->tf_r13 = 0;
117 	tf->tf_r14 = 0;
118 	tf->tf_r15 = 0;
119 	tf->tf_rip = epp->ep_entry;
120 	tf->tf_rflags = PSL_USERSET;
121 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
122 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
123 	tf->tf_ds = 0;
124 	tf->tf_es = 0;
125 	tf->tf_fs = 0;
126 	tf->tf_gs = 0;
127 
128 	return;
129 }
130 
131 void
132 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask)
133 {
134 	struct lwp *l = curlwp;
135 	struct proc *p = l->l_proc;
136 	struct sigacts *ps = p->p_sigacts;
137 	int onstack, error;
138 	int sig = ksi->ksi_signo;
139 	struct linux_rt_sigframe *sfp, sigframe;
140 	struct linux__fpstate *fpsp, fpstate;
141 	struct fpreg fpregs;
142 	struct trapframe *tf = l->l_md.md_regs;
143 	sig_t catcher = SIGACTION(p, sig).sa_handler;
144 	linux_sigset_t lmask;
145 	char *sp;
146 
147 	/* Do we need to jump onto the signal stack? */
148 	onstack =
149 	    (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
150 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
151 
152 	/* Allocate space for the signal handler context. */
153 	if (onstack)
154 		sp = ((char *)l->l_sigstk.ss_sp +
155 		    l->l_sigstk.ss_size);
156 	else
157 		sp = (char *)tf->tf_rsp - 128;
158 
159 	/*
160 	 * Save FPU state, if any
161 	 */
162 	if (l->l_md.md_flags & MDP_USEDFPU) {
163 		sp = (char *)
164 		    (((long)sp - sizeof(struct linux__fpstate)) & ~0xfUL);
165 		fpsp = (struct linux__fpstate *)sp;
166 	} else
167 		fpsp = NULL;
168 
169 	/*
170 	 * Populate the rt_sigframe
171 	 */
172 	sp = (char *)
173 	    ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8);
174 	sfp = (struct linux_rt_sigframe *)sp;
175 
176 	bzero(&sigframe, sizeof(sigframe));
177 	if (ps->sa_sigdesc[sig].sd_vers != 0)
178 		sigframe.pretcode =
179 		    (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp;
180 	else
181 		sigframe.pretcode = NULL;
182 
183 	/*
184 	 * The user context
185 	 */
186 	sigframe.uc.luc_flags = 0;
187 	sigframe.uc.luc_link = NULL;
188 
189 	/* This is used regardless of SA_ONSTACK in Linux */
190 	sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp;
191 	sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size;
192 	sigframe.uc.luc_stack.ss_flags = 0;
193 	if (l->l_sigstk.ss_flags & SS_ONSTACK)
194 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK;
195 	if (l->l_sigstk.ss_flags & SS_DISABLE)
196 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE;
197 
198 	sigframe.uc.luc_mcontext.r8 = tf->tf_r8;
199 	sigframe.uc.luc_mcontext.r9 = tf->tf_r9;
200 	sigframe.uc.luc_mcontext.r10 = tf->tf_r10;
201 	sigframe.uc.luc_mcontext.r11 = tf->tf_r11;
202 	sigframe.uc.luc_mcontext.r12 = tf->tf_r12;
203 	sigframe.uc.luc_mcontext.r13 = tf->tf_r13;
204 	sigframe.uc.luc_mcontext.r14 = tf->tf_r14;
205 	sigframe.uc.luc_mcontext.r15 = tf->tf_r15;
206 	sigframe.uc.luc_mcontext.rdi = tf->tf_rdi;
207 	sigframe.uc.luc_mcontext.rsi = tf->tf_rsi;
208 	sigframe.uc.luc_mcontext.rbp = tf->tf_rbp;
209 	sigframe.uc.luc_mcontext.rbx = tf->tf_rbx;
210 	sigframe.uc.luc_mcontext.rdx = tf->tf_rdx;
211 	sigframe.uc.luc_mcontext.rax = tf->tf_rax;
212 	sigframe.uc.luc_mcontext.rcx = tf->tf_rcx;
213 	sigframe.uc.luc_mcontext.rsp = tf->tf_rsp;
214 	sigframe.uc.luc_mcontext.rip = tf->tf_rip;
215 	sigframe.uc.luc_mcontext.eflags = tf->tf_rflags;
216 	sigframe.uc.luc_mcontext.cs = tf->tf_cs;
217 	sigframe.uc.luc_mcontext.gs = tf->tf_gs;
218 	sigframe.uc.luc_mcontext.fs = tf->tf_fs;
219 	sigframe.uc.luc_mcontext.err = tf->tf_err;
220 	sigframe.uc.luc_mcontext.trapno = tf->tf_trapno;
221 	native_to_linux_sigset(&lmask, mask);
222 	sigframe.uc.luc_mcontext.oldmask = lmask.sig[0];
223 	sigframe.uc.luc_mcontext.cr2 = (long)l->l_addr->u_pcb.pcb_onfault;
224 	sigframe.uc.luc_mcontext.fpstate = fpsp;
225 	native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask);
226 
227 	/*
228 	 * the siginfo structure
229 	 */
230 	sigframe.info.lsi_signo = native_to_linux_signo[sig];
231 	sigframe.info.lsi_errno = native_to_linux_errno[ksi->ksi_errno];
232 	sigframe.info.lsi_code = ksi->ksi_code;
233 
234 	/* XXX This is a rought conversion, taken from i386 code */
235 	switch (sigframe.info.lsi_signo) {
236 	case LINUX_SIGILL:
237 	case LINUX_SIGFPE:
238 	case LINUX_SIGSEGV:
239 	case LINUX_SIGBUS:
240 	case LINUX_SIGTRAP:
241 		sigframe.info._sifields._sigfault._addr = ksi->ksi_addr;
242 		break;
243 	case LINUX_SIGCHLD:
244 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
245 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
246 		sigframe.info._sifields._sigchld._utime = ksi->ksi_utime;
247 		sigframe.info._sifields._sigchld._stime = ksi->ksi_stime;
248 
249 		if (WCOREDUMP(ksi->ksi_status)) {
250 			sigframe.info.lsi_code = LINUX_CLD_DUMPED;
251 			sigframe.info._sifields._sigchld._status =
252 			    _WSTATUS(ksi->ksi_status);
253 		} else if (_WSTATUS(ksi->ksi_status)) {
254 			sigframe.info.lsi_code = LINUX_CLD_KILLED;
255 			sigframe.info._sifields._sigchld._status =
256 			    _WSTATUS(ksi->ksi_status);
257 		} else {
258 			sigframe.info.lsi_code = LINUX_CLD_EXITED;
259 			sigframe.info._sifields._sigchld._status =
260 			    ((ksi->ksi_status & 0xff00U) >> 8);
261 		}
262 		break;
263 	case LINUX_SIGIO:
264 		sigframe.info._sifields._sigpoll._band = ksi->ksi_band;
265 		sigframe.info._sifields._sigpoll._fd = ksi->ksi_fd;
266 		break;
267 	default:
268 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
269 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
270 		if ((sigframe.info.lsi_signo == LINUX_SIGALRM) ||
271 		    (sigframe.info.lsi_signo >= LINUX_SIGRTMIN))
272 			sigframe.info._sifields._timer._sigval.sival_ptr =
273 			     ksi->ksi_value.sival_ptr;
274 		break;
275 	}
276 
277 	sendsig_reset(l, sig);
278 	mutex_exit(&p->p_smutex);
279 	error = 0;
280 
281 	/*
282 	 * Save FPU state, if any
283 	 */
284 	if (fpsp != NULL) {
285 		(void)process_read_fpregs(l, &fpregs);
286 		bzero(&fpstate, sizeof(fpstate));
287 		fpstate.cwd = fpregs.fp_fcw;
288 		fpstate.swd = fpregs.fp_fsw;
289 		fpstate.twd = fpregs.fp_ftw;
290 		fpstate.fop = fpregs.fp_fop;
291 		fpstate.rip = fpregs.fp_rip;
292 		fpstate.rdp = fpregs.fp_rdp;
293 		fpstate.mxcsr = fpregs.fp_mxcsr;
294 		fpstate.mxcsr_mask = fpregs.fp_mxcsr_mask;
295 		memcpy(&fpstate.st_space, &fpregs.fp_st,
296 		    sizeof(fpstate.st_space));
297 		memcpy(&fpstate.xmm_space, &fpregs.fp_xmm,
298 		    sizeof(fpstate.xmm_space));
299 		error = copyout(&fpstate, fpsp, sizeof(fpstate));
300 	}
301 
302 	if (error == 0)
303 		error = copyout(&sigframe, sp, sizeof(sigframe));
304 
305 	mutex_enter(&p->p_smutex);
306 
307 	if (error != 0) {
308 		sigexit(l, SIGILL);
309 		return;
310 	}
311 
312 	linux_buildcontext(l, catcher, sp);
313 	tf->tf_rdi = sigframe.info.lsi_signo;
314 	tf->tf_rax = 0;
315 	tf->tf_rsi = (long)&sfp->info;
316 	tf->tf_rdx = (long)&sfp->uc;
317 
318 	/*
319 	 * Remember we use signal stack
320 	 */
321 	if (onstack)
322 		l->l_sigstk.ss_flags |= SS_ONSTACK;
323 	return;
324 }
325 
326 int
327 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval)
328 {
329 	printf("linux_sys_modify_ldt\n");
330 	return 0;
331 }
332 
333 int
334 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval)
335 {
336 	return 0;
337 }
338 
339 int
340 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval)
341 {
342 	return 0;
343 }
344 
345 dev_t
346 linux_fakedev(dev_t dev, int raw)
347 {
348 
349        extern const struct cdevsw ptc_cdevsw, pts_cdevsw;
350        const struct cdevsw *cd = cdevsw_lookup(dev);
351 
352        if (raw) {
353 #if (NWSDISPLAY > 0)
354 	       extern const struct cdevsw wsdisplay_cdevsw;
355 	       if (cd == &wsdisplay_cdevsw)
356 		       return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
357 #endif
358        }
359 
360        if (cd == &ptc_cdevsw)
361 	       return makedev(LINUX_PTC_MAJOR, minor(dev));
362        if (cd == &pts_cdevsw)
363 	       return makedev(LINUX_PTS_MAJOR, minor(dev));
364 
365 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
366 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
367 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
368 }
369 
370 int
371 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval)
372 {
373 	return 0;
374 }
375 
376 int
377 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval)
378 {
379 	struct linux_ucontext *luctx;
380 	struct trapframe *tf = l->l_md.md_regs;
381 	struct linux_sigcontext *lsigctx;
382 	struct linux__fpstate fpstate;
383 	struct linux_rt_sigframe frame, *fp;
384 	ucontext_t uctx;
385 	mcontext_t *mctx;
386 	struct fxsave64 *fxarea;
387 	int error;
388 
389 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
390 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
391 		mutex_enter(&l->l_proc->p_smutex);
392 		sigexit(l, SIGILL);
393 		return error;
394 	}
395 	luctx = &frame.uc;
396 	lsigctx = &luctx->luc_mcontext;
397 
398 	bzero(&uctx, sizeof(uctx));
399 	mctx = (mcontext_t *)&uctx.uc_mcontext;
400 	fxarea = (struct fxsave64 *)&mctx->__fpregs;
401 
402 	/*
403 	 * Set the flags. Linux always have CPU, stack and signal state,
404 	 * FPU is optional. uc_flags is not used to tell what we have.
405 	 */
406 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
407 	if (lsigctx->fpstate != NULL)
408 		uctx.uc_flags |= _UC_FPU;
409 	uctx.uc_link = NULL;
410 
411 	/*
412 	 * Signal set
413 	 */
414 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
415 
416 	/*
417 	 * CPU state
418 	 */
419 	mctx->__gregs[_REG_R8] = lsigctx->r8;
420 	mctx->__gregs[_REG_R9] = lsigctx->r9;
421 	mctx->__gregs[_REG_R10] = lsigctx->r10;
422 	mctx->__gregs[_REG_R11] = lsigctx->r11;
423 	mctx->__gregs[_REG_R12] = lsigctx->r12;
424 	mctx->__gregs[_REG_R13] = lsigctx->r13;
425 	mctx->__gregs[_REG_R14] = lsigctx->r14;
426 	mctx->__gregs[_REG_R15] = lsigctx->r15;
427 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
428 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
429 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
430 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
431 	mctx->__gregs[_REG_RAX] = lsigctx->rax;
432 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
433 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
434 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
435 	mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags;
436 	mctx->__gregs[_REG_CS] = lsigctx->cs;
437 	mctx->__gregs[_REG_GS] = lsigctx->gs;
438 	mctx->__gregs[_REG_FS] = lsigctx->fs;
439 	mctx->__gregs[_REG_ERR] = lsigctx->err;
440 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
441 	mctx->__gregs[_REG_ES] = tf->tf_es;
442 	mctx->__gregs[_REG_DS] = tf->tf_ds;
443 	mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */
444 	mctx->__gregs[_REG_SS] = tf->tf_ss;
445 
446 	/*
447 	 * FPU state
448 	 */
449 	if (lsigctx->fpstate != NULL) {
450 		error = copyin(lsigctx->fpstate, &fpstate, sizeof(fpstate));
451 		if (error != 0) {
452 			mutex_enter(&l->l_proc->p_smutex);
453 			sigexit(l, SIGILL);
454 			return error;
455 		}
456 
457 		fxarea->fx_fcw = fpstate.cwd;
458 		fxarea->fx_fsw = fpstate.swd;
459 		fxarea->fx_ftw = fpstate.twd;
460 		fxarea->fx_fop = fpstate.fop;
461 		fxarea->fx_rip = fpstate.rip;
462 		fxarea->fx_rdp = fpstate.rdp;
463 		fxarea->fx_mxcsr = fpstate.mxcsr;
464 		fxarea->fx_mxcsr_mask = fpstate.mxcsr_mask;
465 		memcpy(&fxarea->fx_st, &fpstate.st_space,
466 		    sizeof(fxarea->fx_st));
467 		memcpy(&fxarea->fx_xmm, &fpstate.xmm_space,
468 		    sizeof(fxarea->fx_xmm));
469 	}
470 
471 	/*
472 	 * And the stack
473 	 */
474 	uctx.uc_stack.ss_flags = 0;
475 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK);
476 		uctx.uc_stack.ss_flags = SS_ONSTACK;
477 
478 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE);
479 		uctx.uc_stack.ss_flags = SS_DISABLE;
480 
481 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
482 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
483 
484 	/*
485 	 * And let setucontext deal with that.
486 	 */
487 	mutex_enter(&l->l_proc->p_smutex);
488 	error = setucontext(l, &uctx);
489 	mutex_exit(&l->l_proc->p_smutex);
490 	if (error)
491 		return error;
492 
493 	return EJUSTRETURN;
494 }
495 
496 int
497 linux_sys_arch_prctl(struct lwp *l, const struct linux_sys_arch_prctl_args *uap, register_t *retval)
498 {
499 	/* {
500 		syscallarg(int) code;
501 		syscallarg(unsigned long) addr;
502 	} */
503 	struct pcb *pcb = &l->l_addr->u_pcb;
504 	struct trapframe *tf = l->l_md.md_regs;
505 	int error;
506 	uint64_t taddr;
507 
508 	switch(SCARG(uap, code)) {
509 	case LINUX_ARCH_SET_GS:
510 		taddr = SCARG(uap, addr);
511 		if (taddr >= VM_MAXUSER_ADDRESS)
512 			return EINVAL;
513 		pcb->pcb_gs = taddr;
514 		pcb->pcb_flags |= PCB_GS64;
515 		if (l == curlwp)
516 			wrmsr(MSR_KERNELGSBASE, taddr);
517 		break;
518 
519 	case LINUX_ARCH_GET_GS:
520 		if (pcb->pcb_flags & PCB_GS64)
521 			taddr = pcb->pcb_gs;
522 		else {
523 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
524 			if (error != 0)
525 				return error;
526 		}
527 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
528 		if (error != 0)
529 			return error;
530 		break;
531 
532 	case LINUX_ARCH_SET_FS:
533 		taddr = SCARG(uap, addr);
534 		if (taddr >= VM_MAXUSER_ADDRESS)
535 			return EINVAL;
536 		pcb->pcb_fs = taddr;
537 		pcb->pcb_flags |= PCB_FS64;
538 		if (l == curlwp)
539 			wrmsr(MSR_FSBASE, taddr);
540 		break;
541 
542 	case LINUX_ARCH_GET_FS:
543 		if (pcb->pcb_flags & PCB_FS64)
544 			taddr = pcb->pcb_fs;
545 		else {
546 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
547 			if (error != 0)
548 				return error;
549 		}
550 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
551 		if (error != 0)
552 			return error;
553 		break;
554 
555 	default:
556 #ifdef DEBUG_LINUX
557 		printf("linux_sys_arch_prctl: unexpected code %d\n",
558 		    SCARG(uap, code));
559 #endif
560 		return EINVAL;
561 	}
562 
563 	return 0;
564 }
565 
566 const int linux_vsyscall_to_syscall[] = {
567 	LINUX_SYS_gettimeofday,
568 	LINUX_SYS_time,
569 	LINUX_SYS_nosys,
570 	LINUX_SYS_nosys,
571 };
572 
573 int
574 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
575 {
576 	struct trapframe *tf = arg;
577 	uint64_t retaddr;
578 	int vsyscallnr;
579 
580 	/*
581 	 * Check for a vsyscall. %rip must be the fault address,
582 	 * and the address must be in the Linux vsyscall area.
583 	 * Also, vsyscalls are only done at 1024-byte boundaries.
584 	 */
585 
586 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
587 		return 0;
588 
589 	if (trapaddr != tf->tf_rip)
590 		return 0;
591 
592 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
593 		return 0;
594 
595 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
596 
597 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
598 		return 0;
599 
600 	/*
601 	 * Get the return address from the top of the stack,
602 	 * and fix up the return address.
603 	 * This assumes the faulting instruction was callq *reg,
604 	 * which is the only way that vsyscalls are ever entered.
605 	 */
606 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
607 		return 0;
608 	tf->tf_rip = retaddr;
609 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
610 	tf->tf_rsp += 8;	/* "pop" the return address */
611 
612 #if 0
613 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
614 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
615 	    vsyscallnr, (int)tf->tf_rax);
616 #endif
617 
618 	(*l->l_proc->p_md.md_syscall)(tf);
619 
620 	return 1;
621 }
622 
623 static void
624 linux_buildcontext(struct lwp *l, void *catcher, void *f)
625 {
626 	struct trapframe *tf = l->l_md.md_regs;
627 
628 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
629 	tf->tf_rip = (u_int64_t)catcher;
630 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
631 	tf->tf_rflags &= ~(PSL_T|PSL_VM|PSL_AC);
632 	tf->tf_rsp = (u_int64_t)f;
633 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
634 }
635 
636 void *
637 linux_get_newtls(struct lwp *l)
638 {
639 	struct trapframe *tf = l->l_md.md_regs;
640 
641 	return (void *)tf->tf_r8;
642 }
643 
644 int
645 linux_set_newtls(struct lwp *l, void *tls)
646 {
647 	struct linux_sys_arch_prctl_args cup;
648 	register_t retval;
649 
650 	SCARG(&cup, code) = LINUX_ARCH_SET_FS;
651 	SCARG(&cup, addr) = (unsigned long)tls;
652 
653 	return linux_sys_arch_prctl(l, &cup, &retval);
654 }
655