xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision 7c192b2a5e1093666e67801684f930ef49b3b363)
1 /*	$NetBSD: linux_machdep.c,v 1.52 2017/07/14 13:21:29 maxv 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 1.52 2017/07/14 13:21:29 maxv 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/ucontext.h>
46 #include <sys/conf.h>
47 
48 #include <machine/reg.h>
49 #include <machine/pcb.h>
50 #include <machine/mcontext.h>
51 #include <machine/specialreg.h>
52 #include <machine/vmparam.h>
53 #include <machine/cpufunc.h>
54 #include <x86/include/sysarch.h>
55 
56 /*
57  * To see whether wscons is configured (for virtual console ioctl calls).
58  */
59 #if defined(_KERNEL_OPT)
60 #include "wsdisplay.h"
61 #endif
62 #if (NWSDISPLAY > 0)
63 #include <dev/wscons/wsconsio.h>
64 #include <dev/wscons/wsdisplay_usl_io.h>
65 #endif
66 
67 
68 #include <compat/linux/common/linux_signal.h>
69 #include <compat/linux/common/linux_errno.h>
70 #include <compat/linux/common/linux_exec.h>
71 #include <compat/linux/common/linux_ioctl.h>
72 #include <compat/linux/common/linux_prctl.h>
73 #include <compat/linux/common/linux_machdep.h>
74 #include <compat/linux/common/linux_ipc.h>
75 #include <compat/linux/common/linux_sem.h>
76 #include <compat/linux/linux_syscall.h>
77 #include <compat/linux/linux_syscallargs.h>
78 
79 static void linux_buildcontext(struct lwp *, void *, void *);
80 
81 void
82 linux_setregs(struct lwp *l, struct exec_package *epp, vaddr_t stack)
83 {
84 	struct pcb *pcb = lwp_getpcb(l);
85 	struct trapframe *tf;
86 
87 #ifdef USER_LDT
88 	pmap_ldt_cleanup(l);
89 #endif
90 
91 	fpu_save_area_clear(l, __NetBSD_NPXCW__);
92 	pcb->pcb_flags = 0;
93 
94 	l->l_proc->p_flag &= ~PK_32;
95 
96 	l->l_md.md_flags = MDL_IRET;
97 
98 	tf = l->l_md.md_regs;
99 	tf->tf_rax = 0;
100 	tf->tf_rbx = 0;
101 	tf->tf_rcx = epp->ep_entry;
102 	tf->tf_rdx = 0;
103 	tf->tf_rsi = 0;
104 	tf->tf_rdi = 0;
105 	tf->tf_rbp = 0;
106 	tf->tf_rsp = stack;
107 	tf->tf_r8 = 0;
108 	tf->tf_r9 = 0;
109 	tf->tf_r10 = 0;
110 	tf->tf_r11 = 0;
111 	tf->tf_r12 = 0;
112 	tf->tf_r13 = 0;
113 	tf->tf_r14 = 0;
114 	tf->tf_r15 = 0;
115 	tf->tf_rip = epp->ep_entry;
116 	tf->tf_rflags = PSL_USERSET;
117 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
118 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
119 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
120 	tf->tf_es = 0;
121 	cpu_fsgs_zero(l);
122 
123 	return;
124 }
125 
126 void
127 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask)
128 {
129 	struct lwp *l = curlwp;
130 	struct proc *p = l->l_proc;
131 	struct pcb *pcb = lwp_getpcb(l);
132 	struct sigacts *ps = p->p_sigacts;
133 	int onstack, error;
134 	int sig = ksi->ksi_signo;
135 	struct linux_rt_sigframe *sfp, sigframe;
136 	struct linux__fpstate *fpsp;
137 	struct fpreg fpregs;
138 	struct trapframe *tf = l->l_md.md_regs;
139 	sig_t catcher = SIGACTION(p, sig).sa_handler;
140 	linux_sigset_t lmask;
141 	char *sp;
142 
143 	/* Do we need to jump onto the signal stack? */
144 	onstack =
145 	    (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
146 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
147 
148 	/* Allocate space for the signal handler context. */
149 	if (onstack)
150 		sp = ((char *)l->l_sigstk.ss_sp +
151 		    l->l_sigstk.ss_size);
152 	else
153 		sp = (char *)tf->tf_rsp - 128;
154 
155 	/* Save FPU state */
156 	sp = (char *) (((long)sp - sizeof (*fpsp)) & ~0xfUL);
157 	fpsp = (struct linux__fpstate *)sp;
158 
159 	/*
160 	 * Populate the rt_sigframe
161 	 */
162 	sp = (char *)
163 	    ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8);
164 	sfp = (struct linux_rt_sigframe *)sp;
165 
166 	memset(&sigframe, 0, sizeof(sigframe));
167 	if (ps->sa_sigdesc[sig].sd_vers != 0)
168 		sigframe.pretcode =
169 		    (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp;
170 	else
171 		sigframe.pretcode = NULL;
172 
173 	/*
174 	 * The user context
175 	 */
176 	sigframe.uc.luc_flags = 0;
177 	sigframe.uc.luc_link = NULL;
178 
179 	/* This is used regardless of SA_ONSTACK in Linux */
180 	sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp;
181 	sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size;
182 	sigframe.uc.luc_stack.ss_flags = 0;
183 	if (l->l_sigstk.ss_flags & SS_ONSTACK)
184 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK;
185 	if (l->l_sigstk.ss_flags & SS_DISABLE)
186 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE;
187 
188 	sigframe.uc.luc_mcontext.r8 = tf->tf_r8;
189 	sigframe.uc.luc_mcontext.r9 = tf->tf_r9;
190 	sigframe.uc.luc_mcontext.r10 = tf->tf_r10;
191 	sigframe.uc.luc_mcontext.r11 = tf->tf_r11;
192 	sigframe.uc.luc_mcontext.r12 = tf->tf_r12;
193 	sigframe.uc.luc_mcontext.r13 = tf->tf_r13;
194 	sigframe.uc.luc_mcontext.r14 = tf->tf_r14;
195 	sigframe.uc.luc_mcontext.r15 = tf->tf_r15;
196 	sigframe.uc.luc_mcontext.rdi = tf->tf_rdi;
197 	sigframe.uc.luc_mcontext.rsi = tf->tf_rsi;
198 	sigframe.uc.luc_mcontext.rbp = tf->tf_rbp;
199 	sigframe.uc.luc_mcontext.rbx = tf->tf_rbx;
200 	sigframe.uc.luc_mcontext.rdx = tf->tf_rdx;
201 	sigframe.uc.luc_mcontext.rax = tf->tf_rax;
202 	sigframe.uc.luc_mcontext.rcx = tf->tf_rcx;
203 	sigframe.uc.luc_mcontext.rsp = tf->tf_rsp;
204 	sigframe.uc.luc_mcontext.rip = tf->tf_rip;
205 	sigframe.uc.luc_mcontext.eflags = tf->tf_rflags;
206 	sigframe.uc.luc_mcontext.cs = tf->tf_cs;
207 	sigframe.uc.luc_mcontext.gs = tf->tf_gs;
208 	sigframe.uc.luc_mcontext.fs = tf->tf_fs;
209 	sigframe.uc.luc_mcontext.err = tf->tf_err;
210 	sigframe.uc.luc_mcontext.trapno = tf->tf_trapno;
211 	native_to_linux_sigset(&lmask, mask);
212 	sigframe.uc.luc_mcontext.oldmask = lmask.sig[0];
213 	sigframe.uc.luc_mcontext.cr2 = (long)pcb->pcb_onfault;
214 	sigframe.uc.luc_mcontext.fpstate = fpsp;
215 	native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask);
216 	native_to_linux_siginfo(&sigframe.info, &ksi->ksi_info);
217 	sendsig_reset(l, sig);
218 	mutex_exit(p->p_lock);
219 	error = 0;
220 
221 	/*
222 	 * Save FPU state, if any
223 	 */
224 	if (fpsp != NULL) {
225 		size_t fp_size = sizeof fpregs;
226 		/* The netbsd and linux structures both match the fxsave data */
227 		(void)process_read_fpregs(l, &fpregs, &fp_size);
228 		error = copyout(&fpregs, fpsp, sizeof(*fpsp));
229 	}
230 
231 	if (error == 0)
232 		error = copyout(&sigframe, sp, sizeof(sigframe));
233 
234 	mutex_enter(p->p_lock);
235 
236 	if (error != 0) {
237 		sigexit(l, SIGILL);
238 		return;
239 	}
240 
241 	if ((vaddr_t)catcher >= VM_MAXUSER_ADDRESS) {
242 		sigexit(l, SIGILL);
243 		return;
244 	}
245 
246 	linux_buildcontext(l, catcher, sp);
247 	tf->tf_rdi = sigframe.info.lsi_signo;
248 	tf->tf_rax = 0;
249 	tf->tf_rsi = (long)&sfp->info;
250 	tf->tf_rdx = (long)&sfp->uc;
251 
252 	/*
253 	 * Remember we use signal stack
254 	 */
255 	if (onstack)
256 		l->l_sigstk.ss_flags |= SS_ONSTACK;
257 	return;
258 }
259 
260 int
261 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval)
262 {
263 	printf("linux_sys_modify_ldt\n");
264 	return 0;
265 }
266 
267 int
268 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval)
269 {
270 	return 0;
271 }
272 
273 int
274 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval)
275 {
276 	return 0;
277 }
278 
279 dev_t
280 linux_fakedev(dev_t dev, int raw)
281 {
282 
283        extern const struct cdevsw ptc_cdevsw, pts_cdevsw;
284        const struct cdevsw *cd = cdevsw_lookup(dev);
285 
286        if (raw) {
287 #if (NWSDISPLAY > 0)
288 	       extern const struct cdevsw wsdisplay_cdevsw;
289 	       if (cd == &wsdisplay_cdevsw)
290 		       return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
291 #endif
292        }
293 
294        if (cd == &ptc_cdevsw)
295 	       return makedev(LINUX_PTC_MAJOR, minor(dev));
296        if (cd == &pts_cdevsw)
297 	       return makedev(LINUX_PTS_MAJOR, minor(dev));
298 
299 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
300 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
301 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
302 }
303 
304 int
305 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval)
306 {
307 	return 0;
308 }
309 
310 int
311 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval)
312 {
313 	struct linux_ucontext *luctx;
314 	struct trapframe *tf = l->l_md.md_regs;
315 	struct linux_sigcontext *lsigctx;
316 	struct linux_rt_sigframe frame, *fp;
317 	ucontext_t uctx;
318 	mcontext_t *mctx;
319 	struct fxsave *fxarea;
320 	int error;
321 
322 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
323 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
324 		mutex_enter(l->l_proc->p_lock);
325 		sigexit(l, SIGILL);
326 		return error;
327 	}
328 	luctx = &frame.uc;
329 	lsigctx = &luctx->luc_mcontext;
330 
331 	memset(&uctx, 0, sizeof(uctx));
332 	mctx = (mcontext_t *)&uctx.uc_mcontext;
333 	fxarea = (struct fxsave *)&mctx->__fpregs;
334 
335 	/*
336 	 * Set the flags. Linux always have CPU, stack and signal state,
337 	 * FPU is optional. uc_flags is not used to tell what we have.
338 	 */
339 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
340 	if (lsigctx->fpstate != NULL)
341 		uctx.uc_flags |= _UC_FPU;
342 	uctx.uc_link = NULL;
343 
344 	/*
345 	 * Signal set
346 	 */
347 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
348 
349 	/*
350 	 * CPU state
351 	 */
352 	mctx->__gregs[_REG_R8] = lsigctx->r8;
353 	mctx->__gregs[_REG_R9] = lsigctx->r9;
354 	mctx->__gregs[_REG_R10] = lsigctx->r10;
355 	mctx->__gregs[_REG_R11] = lsigctx->r11;
356 	mctx->__gregs[_REG_R12] = lsigctx->r12;
357 	mctx->__gregs[_REG_R13] = lsigctx->r13;
358 	mctx->__gregs[_REG_R14] = lsigctx->r14;
359 	mctx->__gregs[_REG_R15] = lsigctx->r15;
360 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
361 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
362 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
363 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
364 	mctx->__gregs[_REG_RAX] = lsigctx->rax;
365 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
366 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
367 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
368 	mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags;
369 	mctx->__gregs[_REG_CS] = lsigctx->cs;
370 	mctx->__gregs[_REG_GS] = lsigctx->gs;
371 	mctx->__gregs[_REG_FS] = lsigctx->fs;
372 	mctx->__gregs[_REG_ERR] = lsigctx->err;
373 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
374 	mctx->__gregs[_REG_ES] = tf->tf_es;
375 	mctx->__gregs[_REG_DS] = tf->tf_ds;
376 	mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */
377 	mctx->__gregs[_REG_SS] = tf->tf_ss;
378 
379 	/*
380 	 * FPU state
381 	 */
382 	if (lsigctx->fpstate != NULL) {
383 		/* Both structures match the fxstate data */
384 		error = copyin(lsigctx->fpstate, fxarea, sizeof(*fxarea));
385 		if (error != 0) {
386 			mutex_enter(l->l_proc->p_lock);
387 			sigexit(l, SIGILL);
388 			return error;
389 		}
390 	}
391 
392 	/*
393 	 * And the stack
394 	 */
395 	uctx.uc_stack.ss_flags = 0;
396 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK)
397 		uctx.uc_stack.ss_flags |= SS_ONSTACK;
398 
399 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE)
400 		uctx.uc_stack.ss_flags |= SS_DISABLE;
401 
402 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
403 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
404 
405 	/*
406 	 * And let setucontext deal with that.
407 	 */
408 	mutex_enter(l->l_proc->p_lock);
409 	error = setucontext(l, &uctx);
410 	mutex_exit(l->l_proc->p_lock);
411 	if (error)
412 		return error;
413 
414 	return EJUSTRETURN;
415 }
416 
417 int
418 linux_sys_arch_prctl(struct lwp *l,
419     const struct linux_sys_arch_prctl_args *uap, register_t *retval)
420 {
421 	/* {
422 		syscallarg(int) code;
423 		syscallarg(unsigned long) addr;
424 	} */
425 	void *addr = (void *)SCARG(uap, addr);
426 
427 	switch(SCARG(uap, code)) {
428 	case LINUX_ARCH_SET_GS:
429 		return x86_set_sdbase(addr, 'g', l, true);
430 
431 	case LINUX_ARCH_GET_GS:
432 		return x86_get_sdbase(addr, 'g');
433 
434 	case LINUX_ARCH_SET_FS:
435 		return x86_set_sdbase(addr, 'f', l, true);
436 
437 	case LINUX_ARCH_GET_FS:
438 		return x86_get_sdbase(addr, 'f');
439 
440 	default:
441 #ifdef DEBUG_LINUX
442 		printf("linux_sys_arch_prctl: unexpected code %d\n",
443 		    SCARG(uap, code));
444 #endif
445 		return EINVAL;
446 	}
447 	/* NOTREACHED */
448 }
449 
450 const int linux_vsyscall_to_syscall[] = {
451 	LINUX_SYS_gettimeofday,
452 	LINUX_SYS_time,
453 	LINUX_SYS_nosys,	/* nosys */
454 	LINUX_SYS_nosys,	/* nosys */
455 };
456 
457 int
458 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
459 {
460 	struct trapframe *tf = arg;
461 	uint64_t retaddr;
462 	size_t vsyscallnr;
463 
464 	/*
465 	 * Check for a vsyscall. %rip must be the fault address,
466 	 * and the address must be in the Linux vsyscall area.
467 	 * Also, vsyscalls are only done at 1024-byte boundaries.
468 	 */
469 
470 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
471 		return 0;
472 
473 	if (trapaddr != tf->tf_rip)
474 		return 0;
475 
476 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
477 		return 0;
478 
479 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
480 
481 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
482 		return 0;
483 
484 	/*
485 	 * Get the return address from the top of the stack,
486 	 * and fix up the return address.
487 	 * This assumes the faulting instruction was callq *reg,
488 	 * which is the only way that vsyscalls are ever entered.
489 	 */
490 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
491 		return 0;
492 	if ((vaddr_t)retaddr >= VM_MAXUSER_ADDRESS)
493 		return 0;
494 	tf->tf_rip = retaddr;
495 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
496 	tf->tf_rsp += 8;	/* "pop" the return address */
497 
498 #if 0
499 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
500 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
501 	    vsyscallnr, (int)tf->tf_rax);
502 #endif
503 
504 	(*l->l_proc->p_md.md_syscall)(tf);
505 
506 	return 1;
507 }
508 
509 static void
510 linux_buildcontext(struct lwp *l, void *catcher, void *f)
511 {
512 	struct trapframe *tf = l->l_md.md_regs;
513 
514 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
515 	tf->tf_rip = (u_int64_t)catcher;
516 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
517 	tf->tf_rflags &= ~PSL_CLEARSIG;
518 	tf->tf_rsp = (u_int64_t)f;
519 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
520 }
521