xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision da9817918ec7e88db2912a2882967c7570a83f47)
1 /*	$NetBSD: linux_machdep.c,v 1.36 2009/05/29 14:19:12 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 1.36 2009/05/29 14:19:12 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/ucontext.h>
47 #include <sys/conf.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 #include <machine/cpufunc.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 	memset(&sigframe, 0, 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 = native_to_linux_si_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 		sigframe.info._sifields._sigchld._status =
249 		    native_to_linux_si_status(ksi->ksi_code, ksi->ksi_status);
250 		break;
251 	case LINUX_SIGIO:
252 		sigframe.info._sifields._sigpoll._band = ksi->ksi_band;
253 		sigframe.info._sifields._sigpoll._fd = ksi->ksi_fd;
254 		break;
255 	default:
256 		sigframe.info._sifields._sigchld._pid = ksi->ksi_pid;
257 		sigframe.info._sifields._sigchld._uid = ksi->ksi_uid;
258 		if ((sigframe.info.lsi_signo == LINUX_SIGALRM) ||
259 		    (sigframe.info.lsi_signo >= LINUX_SIGRTMIN))
260 			sigframe.info._sifields._timer._sigval.sival_ptr =
261 			     ksi->ksi_value.sival_ptr;
262 		break;
263 	}
264 
265 	sendsig_reset(l, sig);
266 	mutex_exit(p->p_lock);
267 	error = 0;
268 
269 	/*
270 	 * Save FPU state, if any
271 	 */
272 	if (fpsp != NULL) {
273 		(void)process_read_fpregs(l, &fpregs);
274 		memset(&fpstate, 0, sizeof(fpstate));
275 		fpstate.cwd = fpregs.fp_fcw;
276 		fpstate.swd = fpregs.fp_fsw;
277 		fpstate.twd = fpregs.fp_ftw;
278 		fpstate.fop = fpregs.fp_fop;
279 		fpstate.rip = fpregs.fp_rip;
280 		fpstate.rdp = fpregs.fp_rdp;
281 		fpstate.mxcsr = fpregs.fp_mxcsr;
282 		fpstate.mxcsr_mask = fpregs.fp_mxcsr_mask;
283 		memcpy(&fpstate.st_space, &fpregs.fp_st,
284 		    sizeof(fpstate.st_space));
285 		memcpy(&fpstate.xmm_space, &fpregs.fp_xmm,
286 		    sizeof(fpstate.xmm_space));
287 		error = copyout(&fpstate, fpsp, sizeof(fpstate));
288 	}
289 
290 	if (error == 0)
291 		error = copyout(&sigframe, sp, sizeof(sigframe));
292 
293 	mutex_enter(p->p_lock);
294 
295 	if (error != 0) {
296 		sigexit(l, SIGILL);
297 		return;
298 	}
299 
300 	linux_buildcontext(l, catcher, sp);
301 	tf->tf_rdi = sigframe.info.lsi_signo;
302 	tf->tf_rax = 0;
303 	tf->tf_rsi = (long)&sfp->info;
304 	tf->tf_rdx = (long)&sfp->uc;
305 
306 	/*
307 	 * Remember we use signal stack
308 	 */
309 	if (onstack)
310 		l->l_sigstk.ss_flags |= SS_ONSTACK;
311 	return;
312 }
313 
314 int
315 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval)
316 {
317 	printf("linux_sys_modify_ldt\n");
318 	return 0;
319 }
320 
321 int
322 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval)
323 {
324 	return 0;
325 }
326 
327 int
328 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval)
329 {
330 	return 0;
331 }
332 
333 dev_t
334 linux_fakedev(dev_t dev, int raw)
335 {
336 
337        extern const struct cdevsw ptc_cdevsw, pts_cdevsw;
338        const struct cdevsw *cd = cdevsw_lookup(dev);
339 
340        if (raw) {
341 #if (NWSDISPLAY > 0)
342 	       extern const struct cdevsw wsdisplay_cdevsw;
343 	       if (cd == &wsdisplay_cdevsw)
344 		       return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
345 #endif
346        }
347 
348        if (cd == &ptc_cdevsw)
349 	       return makedev(LINUX_PTC_MAJOR, minor(dev));
350        if (cd == &pts_cdevsw)
351 	       return makedev(LINUX_PTS_MAJOR, minor(dev));
352 
353 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
354 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
355 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
356 }
357 
358 int
359 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval)
360 {
361 	return 0;
362 }
363 
364 int
365 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval)
366 {
367 	struct linux_ucontext *luctx;
368 	struct trapframe *tf = l->l_md.md_regs;
369 	struct linux_sigcontext *lsigctx;
370 	struct linux__fpstate fpstate;
371 	struct linux_rt_sigframe frame, *fp;
372 	ucontext_t uctx;
373 	mcontext_t *mctx;
374 	struct fxsave64 *fxarea;
375 	int error;
376 
377 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
378 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
379 		mutex_enter(l->l_proc->p_lock);
380 		sigexit(l, SIGILL);
381 		return error;
382 	}
383 	luctx = &frame.uc;
384 	lsigctx = &luctx->luc_mcontext;
385 
386 	memset(&uctx, 0, sizeof(uctx));
387 	mctx = (mcontext_t *)&uctx.uc_mcontext;
388 	fxarea = (struct fxsave64 *)&mctx->__fpregs;
389 
390 	/*
391 	 * Set the flags. Linux always have CPU, stack and signal state,
392 	 * FPU is optional. uc_flags is not used to tell what we have.
393 	 */
394 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
395 	if (lsigctx->fpstate != NULL)
396 		uctx.uc_flags |= _UC_FPU;
397 	uctx.uc_link = NULL;
398 
399 	/*
400 	 * Signal set
401 	 */
402 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
403 
404 	/*
405 	 * CPU state
406 	 */
407 	mctx->__gregs[_REG_R8] = lsigctx->r8;
408 	mctx->__gregs[_REG_R9] = lsigctx->r9;
409 	mctx->__gregs[_REG_R10] = lsigctx->r10;
410 	mctx->__gregs[_REG_R11] = lsigctx->r11;
411 	mctx->__gregs[_REG_R12] = lsigctx->r12;
412 	mctx->__gregs[_REG_R13] = lsigctx->r13;
413 	mctx->__gregs[_REG_R14] = lsigctx->r14;
414 	mctx->__gregs[_REG_R15] = lsigctx->r15;
415 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
416 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
417 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
418 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
419 	mctx->__gregs[_REG_RAX] = lsigctx->rax;
420 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
421 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
422 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
423 	mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags;
424 	mctx->__gregs[_REG_CS] = lsigctx->cs;
425 	mctx->__gregs[_REG_GS] = lsigctx->gs;
426 	mctx->__gregs[_REG_FS] = lsigctx->fs;
427 	mctx->__gregs[_REG_ERR] = lsigctx->err;
428 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
429 	mctx->__gregs[_REG_ES] = tf->tf_es;
430 	mctx->__gregs[_REG_DS] = tf->tf_ds;
431 	mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */
432 	mctx->__gregs[_REG_SS] = tf->tf_ss;
433 
434 	/*
435 	 * FPU state
436 	 */
437 	if (lsigctx->fpstate != NULL) {
438 		error = copyin(lsigctx->fpstate, &fpstate, sizeof(fpstate));
439 		if (error != 0) {
440 			mutex_enter(l->l_proc->p_lock);
441 			sigexit(l, SIGILL);
442 			return error;
443 		}
444 
445 		fxarea->fx_fcw = fpstate.cwd;
446 		fxarea->fx_fsw = fpstate.swd;
447 		fxarea->fx_ftw = fpstate.twd;
448 		fxarea->fx_fop = fpstate.fop;
449 		fxarea->fx_rip = fpstate.rip;
450 		fxarea->fx_rdp = fpstate.rdp;
451 		fxarea->fx_mxcsr = fpstate.mxcsr;
452 		fxarea->fx_mxcsr_mask = fpstate.mxcsr_mask;
453 		memcpy(&fxarea->fx_st, &fpstate.st_space,
454 		    sizeof(fxarea->fx_st));
455 		memcpy(&fxarea->fx_xmm, &fpstate.xmm_space,
456 		    sizeof(fxarea->fx_xmm));
457 	}
458 
459 	/*
460 	 * And the stack
461 	 */
462 	uctx.uc_stack.ss_flags = 0;
463 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK)
464 		uctx.uc_stack.ss_flags |= SS_ONSTACK;
465 
466 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE)
467 		uctx.uc_stack.ss_flags |= SS_DISABLE;
468 
469 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
470 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
471 
472 	/*
473 	 * And let setucontext deal with that.
474 	 */
475 	mutex_enter(l->l_proc->p_lock);
476 	error = setucontext(l, &uctx);
477 	mutex_exit(l->l_proc->p_lock);
478 	if (error)
479 		return error;
480 
481 	return EJUSTRETURN;
482 }
483 
484 int
485 linux_sys_arch_prctl(struct lwp *l, const struct linux_sys_arch_prctl_args *uap, register_t *retval)
486 {
487 	/* {
488 		syscallarg(int) code;
489 		syscallarg(unsigned long) addr;
490 	} */
491 	struct pcb *pcb = &l->l_addr->u_pcb;
492 	struct trapframe *tf = l->l_md.md_regs;
493 	int error;
494 	uint64_t taddr;
495 
496 	switch(SCARG(uap, code)) {
497 	case LINUX_ARCH_SET_GS:
498 		taddr = SCARG(uap, addr);
499 		if (taddr >= VM_MAXUSER_ADDRESS)
500 			return EINVAL;
501 		pcb->pcb_gs = taddr;
502 		pcb->pcb_flags |= PCB_GS64;
503 		if (l == curlwp)
504 			wrmsr(MSR_KERNELGSBASE, taddr);
505 		break;
506 
507 	case LINUX_ARCH_GET_GS:
508 		if (pcb->pcb_flags & PCB_GS64)
509 			taddr = pcb->pcb_gs;
510 		else {
511 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
512 			if (error != 0)
513 				return error;
514 		}
515 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
516 		if (error != 0)
517 			return error;
518 		break;
519 
520 	case LINUX_ARCH_SET_FS:
521 		taddr = SCARG(uap, addr);
522 		if (taddr >= VM_MAXUSER_ADDRESS)
523 			return EINVAL;
524 		pcb->pcb_fs = taddr;
525 		pcb->pcb_flags |= PCB_FS64;
526 		if (l == curlwp)
527 			wrmsr(MSR_FSBASE, taddr);
528 		break;
529 
530 	case LINUX_ARCH_GET_FS:
531 		if (pcb->pcb_flags & PCB_FS64)
532 			taddr = pcb->pcb_fs;
533 		else {
534 			error = memseg_baseaddr(l, tf->tf_fs, NULL, 0, &taddr);
535 			if (error != 0)
536 				return error;
537 		}
538 		error = copyout(&taddr, (char *)SCARG(uap, addr), 8);
539 		if (error != 0)
540 			return error;
541 		break;
542 
543 	default:
544 #ifdef DEBUG_LINUX
545 		printf("linux_sys_arch_prctl: unexpected code %d\n",
546 		    SCARG(uap, code));
547 #endif
548 		return EINVAL;
549 	}
550 
551 	return 0;
552 }
553 
554 const int linux_vsyscall_to_syscall[] = {
555 	LINUX_SYS_gettimeofday,
556 	LINUX_SYS_time,
557 	LINUX_SYS_nosys,	/* nosys */
558 	LINUX_SYS_nosys,	/* nosys */
559 };
560 
561 int
562 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
563 {
564 	struct trapframe *tf = arg;
565 	uint64_t retaddr;
566 	int vsyscallnr;
567 
568 	/*
569 	 * Check for a vsyscall. %rip must be the fault address,
570 	 * and the address must be in the Linux vsyscall area.
571 	 * Also, vsyscalls are only done at 1024-byte boundaries.
572 	 */
573 
574 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
575 		return 0;
576 
577 	if (trapaddr != tf->tf_rip)
578 		return 0;
579 
580 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
581 		return 0;
582 
583 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
584 
585 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
586 		return 0;
587 
588 	/*
589 	 * Get the return address from the top of the stack,
590 	 * and fix up the return address.
591 	 * This assumes the faulting instruction was callq *reg,
592 	 * which is the only way that vsyscalls are ever entered.
593 	 */
594 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
595 		return 0;
596 	tf->tf_rip = retaddr;
597 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
598 	tf->tf_rsp += 8;	/* "pop" the return address */
599 
600 #if 0
601 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
602 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
603 	    vsyscallnr, (int)tf->tf_rax);
604 #endif
605 
606 	(*l->l_proc->p_md.md_syscall)(tf);
607 
608 	return 1;
609 }
610 
611 static void
612 linux_buildcontext(struct lwp *l, void *catcher, void *f)
613 {
614 	struct trapframe *tf = l->l_md.md_regs;
615 
616 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
617 	tf->tf_rip = (u_int64_t)catcher;
618 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
619 	tf->tf_rflags &= ~PSL_CLEARSIG;
620 	tf->tf_rsp = (u_int64_t)f;
621 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
622 }
623 
624 void *
625 linux_get_newtls(struct lwp *l)
626 {
627 	struct trapframe *tf = l->l_md.md_regs;
628 
629 	return (void *)tf->tf_r8;
630 }
631 
632 int
633 linux_set_newtls(struct lwp *l, void *tls)
634 {
635 	struct linux_sys_arch_prctl_args cup;
636 	register_t retval;
637 
638 	SCARG(&cup, code) = LINUX_ARCH_SET_FS;
639 	SCARG(&cup, addr) = (unsigned long)tls;
640 
641 	return linux_sys_arch_prctl(l, &cup, &retval);
642 }
643