xref: /netbsd-src/sys/compat/linux/arch/amd64/linux_machdep.c (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /*	$NetBSD: linux_machdep.c,v 1.51 2017/02/13 15:03:18 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 1.51 2017/02/13 15:03:18 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 	tf = l->l_md.md_regs;
97 	tf->tf_rax = 0;
98 	tf->tf_rbx = 0;
99 	tf->tf_rcx = epp->ep_entry;
100 	tf->tf_rdx = 0;
101 	tf->tf_rsi = 0;
102 	tf->tf_rdi = 0;
103 	tf->tf_rbp = 0;
104 	tf->tf_rsp = stack;
105 	tf->tf_r8 = 0;
106 	tf->tf_r9 = 0;
107 	tf->tf_r10 = 0;
108 	tf->tf_r11 = 0;
109 	tf->tf_r12 = 0;
110 	tf->tf_r13 = 0;
111 	tf->tf_r14 = 0;
112 	tf->tf_r15 = 0;
113 	tf->tf_rip = epp->ep_entry;
114 	tf->tf_rflags = PSL_USERSET;
115 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
116 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
117 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
118 	tf->tf_es = 0;
119 	cpu_fsgs_zero(l);
120 
121 	return;
122 }
123 
124 void
125 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask)
126 {
127 	struct lwp *l = curlwp;
128 	struct proc *p = l->l_proc;
129 	struct pcb *pcb = lwp_getpcb(l);
130 	struct sigacts *ps = p->p_sigacts;
131 	int onstack, error;
132 	int sig = ksi->ksi_signo;
133 	struct linux_rt_sigframe *sfp, sigframe;
134 	struct linux__fpstate *fpsp;
135 	struct fpreg fpregs;
136 	struct trapframe *tf = l->l_md.md_regs;
137 	sig_t catcher = SIGACTION(p, sig).sa_handler;
138 	linux_sigset_t lmask;
139 	char *sp;
140 
141 	/* Do we need to jump onto the signal stack? */
142 	onstack =
143 	    (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
144 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
145 
146 	/* Allocate space for the signal handler context. */
147 	if (onstack)
148 		sp = ((char *)l->l_sigstk.ss_sp +
149 		    l->l_sigstk.ss_size);
150 	else
151 		sp = (char *)tf->tf_rsp - 128;
152 
153 	/* Save FPU state */
154 	sp = (char *) (((long)sp - sizeof (*fpsp)) & ~0xfUL);
155 	fpsp = (struct linux__fpstate *)sp;
156 
157 	/*
158 	 * Populate the rt_sigframe
159 	 */
160 	sp = (char *)
161 	    ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8);
162 	sfp = (struct linux_rt_sigframe *)sp;
163 
164 	memset(&sigframe, 0, sizeof(sigframe));
165 	if (ps->sa_sigdesc[sig].sd_vers != 0)
166 		sigframe.pretcode =
167 		    (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp;
168 	else
169 		sigframe.pretcode = NULL;
170 
171 	/*
172 	 * The user context
173 	 */
174 	sigframe.uc.luc_flags = 0;
175 	sigframe.uc.luc_link = NULL;
176 
177 	/* This is used regardless of SA_ONSTACK in Linux */
178 	sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp;
179 	sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size;
180 	sigframe.uc.luc_stack.ss_flags = 0;
181 	if (l->l_sigstk.ss_flags & SS_ONSTACK)
182 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK;
183 	if (l->l_sigstk.ss_flags & SS_DISABLE)
184 		sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE;
185 
186 	sigframe.uc.luc_mcontext.r8 = tf->tf_r8;
187 	sigframe.uc.luc_mcontext.r9 = tf->tf_r9;
188 	sigframe.uc.luc_mcontext.r10 = tf->tf_r10;
189 	sigframe.uc.luc_mcontext.r11 = tf->tf_r11;
190 	sigframe.uc.luc_mcontext.r12 = tf->tf_r12;
191 	sigframe.uc.luc_mcontext.r13 = tf->tf_r13;
192 	sigframe.uc.luc_mcontext.r14 = tf->tf_r14;
193 	sigframe.uc.luc_mcontext.r15 = tf->tf_r15;
194 	sigframe.uc.luc_mcontext.rdi = tf->tf_rdi;
195 	sigframe.uc.luc_mcontext.rsi = tf->tf_rsi;
196 	sigframe.uc.luc_mcontext.rbp = tf->tf_rbp;
197 	sigframe.uc.luc_mcontext.rbx = tf->tf_rbx;
198 	sigframe.uc.luc_mcontext.rdx = tf->tf_rdx;
199 	sigframe.uc.luc_mcontext.rax = tf->tf_rax;
200 	sigframe.uc.luc_mcontext.rcx = tf->tf_rcx;
201 	sigframe.uc.luc_mcontext.rsp = tf->tf_rsp;
202 	sigframe.uc.luc_mcontext.rip = tf->tf_rip;
203 	sigframe.uc.luc_mcontext.eflags = tf->tf_rflags;
204 	sigframe.uc.luc_mcontext.cs = tf->tf_cs;
205 	sigframe.uc.luc_mcontext.gs = tf->tf_gs;
206 	sigframe.uc.luc_mcontext.fs = tf->tf_fs;
207 	sigframe.uc.luc_mcontext.err = tf->tf_err;
208 	sigframe.uc.luc_mcontext.trapno = tf->tf_trapno;
209 	native_to_linux_sigset(&lmask, mask);
210 	sigframe.uc.luc_mcontext.oldmask = lmask.sig[0];
211 	sigframe.uc.luc_mcontext.cr2 = (long)pcb->pcb_onfault;
212 	sigframe.uc.luc_mcontext.fpstate = fpsp;
213 	native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask);
214 	native_to_linux_siginfo(&sigframe.info, &ksi->ksi_info);
215 	sendsig_reset(l, sig);
216 	mutex_exit(p->p_lock);
217 	error = 0;
218 
219 	/*
220 	 * Save FPU state, if any
221 	 */
222 	if (fpsp != NULL) {
223 		size_t fp_size = sizeof fpregs;
224 		/* The netbsd and linux structures both match the fxsave data */
225 		(void)process_read_fpregs(l, &fpregs, &fp_size);
226 		error = copyout(&fpregs, fpsp, sizeof(*fpsp));
227 	}
228 
229 	if (error == 0)
230 		error = copyout(&sigframe, sp, sizeof(sigframe));
231 
232 	mutex_enter(p->p_lock);
233 
234 	if (error != 0) {
235 		sigexit(l, SIGILL);
236 		return;
237 	}
238 
239 	if ((vaddr_t)catcher >= VM_MAXUSER_ADDRESS) {
240 		sigexit(l, SIGILL);
241 		return;
242 	}
243 
244 	linux_buildcontext(l, catcher, sp);
245 	tf->tf_rdi = sigframe.info.lsi_signo;
246 	tf->tf_rax = 0;
247 	tf->tf_rsi = (long)&sfp->info;
248 	tf->tf_rdx = (long)&sfp->uc;
249 
250 	/*
251 	 * Remember we use signal stack
252 	 */
253 	if (onstack)
254 		l->l_sigstk.ss_flags |= SS_ONSTACK;
255 	return;
256 }
257 
258 int
259 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval)
260 {
261 	printf("linux_sys_modify_ldt\n");
262 	return 0;
263 }
264 
265 int
266 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval)
267 {
268 	return 0;
269 }
270 
271 int
272 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval)
273 {
274 	return 0;
275 }
276 
277 dev_t
278 linux_fakedev(dev_t dev, int raw)
279 {
280 
281        extern const struct cdevsw ptc_cdevsw, pts_cdevsw;
282        const struct cdevsw *cd = cdevsw_lookup(dev);
283 
284        if (raw) {
285 #if (NWSDISPLAY > 0)
286 	       extern const struct cdevsw wsdisplay_cdevsw;
287 	       if (cd == &wsdisplay_cdevsw)
288 		       return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
289 #endif
290        }
291 
292        if (cd == &ptc_cdevsw)
293 	       return makedev(LINUX_PTC_MAJOR, minor(dev));
294        if (cd == &pts_cdevsw)
295 	       return makedev(LINUX_PTS_MAJOR, minor(dev));
296 
297 	return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8)
298 	    | (((unsigned long long int) (minor(dev) & ~0xff)) << 12)
299 	    | (((unsigned long long int) (major(dev) & ~0xfff)) << 32));
300 }
301 
302 int
303 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval)
304 {
305 	return 0;
306 }
307 
308 int
309 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval)
310 {
311 	struct linux_ucontext *luctx;
312 	struct trapframe *tf = l->l_md.md_regs;
313 	struct linux_sigcontext *lsigctx;
314 	struct linux_rt_sigframe frame, *fp;
315 	ucontext_t uctx;
316 	mcontext_t *mctx;
317 	struct fxsave *fxarea;
318 	int error;
319 
320 	fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8);
321 	if ((error = copyin(fp, &frame, sizeof(frame))) != 0) {
322 		mutex_enter(l->l_proc->p_lock);
323 		sigexit(l, SIGILL);
324 		return error;
325 	}
326 	luctx = &frame.uc;
327 	lsigctx = &luctx->luc_mcontext;
328 
329 	memset(&uctx, 0, sizeof(uctx));
330 	mctx = (mcontext_t *)&uctx.uc_mcontext;
331 	fxarea = (struct fxsave *)&mctx->__fpregs;
332 
333 	/*
334 	 * Set the flags. Linux always have CPU, stack and signal state,
335 	 * FPU is optional. uc_flags is not used to tell what we have.
336 	 */
337 	uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK);
338 	if (lsigctx->fpstate != NULL)
339 		uctx.uc_flags |= _UC_FPU;
340 	uctx.uc_link = NULL;
341 
342 	/*
343 	 * Signal set
344 	 */
345 	linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask);
346 
347 	/*
348 	 * CPU state
349 	 */
350 	mctx->__gregs[_REG_R8] = lsigctx->r8;
351 	mctx->__gregs[_REG_R9] = lsigctx->r9;
352 	mctx->__gregs[_REG_R10] = lsigctx->r10;
353 	mctx->__gregs[_REG_R11] = lsigctx->r11;
354 	mctx->__gregs[_REG_R12] = lsigctx->r12;
355 	mctx->__gregs[_REG_R13] = lsigctx->r13;
356 	mctx->__gregs[_REG_R14] = lsigctx->r14;
357 	mctx->__gregs[_REG_R15] = lsigctx->r15;
358 	mctx->__gregs[_REG_RDI] = lsigctx->rdi;
359 	mctx->__gregs[_REG_RSI] = lsigctx->rsi;
360 	mctx->__gregs[_REG_RBP] = lsigctx->rbp;
361 	mctx->__gregs[_REG_RBX] = lsigctx->rbx;
362 	mctx->__gregs[_REG_RAX] = lsigctx->rax;
363 	mctx->__gregs[_REG_RDX] = lsigctx->rdx;
364 	mctx->__gregs[_REG_RCX] = lsigctx->rcx;
365 	mctx->__gregs[_REG_RIP] = lsigctx->rip;
366 	mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags;
367 	mctx->__gregs[_REG_CS] = lsigctx->cs;
368 	mctx->__gregs[_REG_GS] = lsigctx->gs;
369 	mctx->__gregs[_REG_FS] = lsigctx->fs;
370 	mctx->__gregs[_REG_ERR] = lsigctx->err;
371 	mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno;
372 	mctx->__gregs[_REG_ES] = tf->tf_es;
373 	mctx->__gregs[_REG_DS] = tf->tf_ds;
374 	mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */
375 	mctx->__gregs[_REG_SS] = tf->tf_ss;
376 
377 	/*
378 	 * FPU state
379 	 */
380 	if (lsigctx->fpstate != NULL) {
381 		/* Both structures match the fxstate data */
382 		error = copyin(lsigctx->fpstate, fxarea, sizeof(*fxarea));
383 		if (error != 0) {
384 			mutex_enter(l->l_proc->p_lock);
385 			sigexit(l, SIGILL);
386 			return error;
387 		}
388 	}
389 
390 	/*
391 	 * And the stack
392 	 */
393 	uctx.uc_stack.ss_flags = 0;
394 	if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK)
395 		uctx.uc_stack.ss_flags |= SS_ONSTACK;
396 
397 	if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE)
398 		uctx.uc_stack.ss_flags |= SS_DISABLE;
399 
400 	uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp;
401 	uctx.uc_stack.ss_size = luctx->luc_stack.ss_size;
402 
403 	/*
404 	 * And let setucontext deal with that.
405 	 */
406 	mutex_enter(l->l_proc->p_lock);
407 	error = setucontext(l, &uctx);
408 	mutex_exit(l->l_proc->p_lock);
409 	if (error)
410 		return error;
411 
412 	return EJUSTRETURN;
413 }
414 
415 int
416 linux_sys_arch_prctl(struct lwp *l,
417     const struct linux_sys_arch_prctl_args *uap, register_t *retval)
418 {
419 	/* {
420 		syscallarg(int) code;
421 		syscallarg(unsigned long) addr;
422 	} */
423 	void *addr = (void *)SCARG(uap, addr);
424 
425 	switch(SCARG(uap, code)) {
426 	case LINUX_ARCH_SET_GS:
427 		return x86_set_sdbase(addr, 'g', l, true);
428 
429 	case LINUX_ARCH_GET_GS:
430 		return x86_get_sdbase(addr, 'g');
431 
432 	case LINUX_ARCH_SET_FS:
433 		return x86_set_sdbase(addr, 'f', l, true);
434 
435 	case LINUX_ARCH_GET_FS:
436 		return x86_get_sdbase(addr, 'f');
437 
438 	default:
439 #ifdef DEBUG_LINUX
440 		printf("linux_sys_arch_prctl: unexpected code %d\n",
441 		    SCARG(uap, code));
442 #endif
443 		return EINVAL;
444 	}
445 	/* NOTREACHED */
446 }
447 
448 const int linux_vsyscall_to_syscall[] = {
449 	LINUX_SYS_gettimeofday,
450 	LINUX_SYS_time,
451 	LINUX_SYS_nosys,	/* nosys */
452 	LINUX_SYS_nosys,	/* nosys */
453 };
454 
455 int
456 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
457 {
458 	struct trapframe *tf = arg;
459 	uint64_t retaddr;
460 	size_t vsyscallnr;
461 
462 	/*
463 	 * Check for a vsyscall. %rip must be the fault address,
464 	 * and the address must be in the Linux vsyscall area.
465 	 * Also, vsyscalls are only done at 1024-byte boundaries.
466 	 */
467 
468 	if (__predict_true(trapaddr < LINUX_VSYSCALL_START))
469 		return 0;
470 
471 	if (trapaddr != tf->tf_rip)
472 		return 0;
473 
474 	if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0)
475 		return 0;
476 
477 	vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE;
478 
479 	if (vsyscallnr > LINUX_VSYSCALL_MAXNR)
480 		return 0;
481 
482 	/*
483 	 * Get the return address from the top of the stack,
484 	 * and fix up the return address.
485 	 * This assumes the faulting instruction was callq *reg,
486 	 * which is the only way that vsyscalls are ever entered.
487 	 */
488 	if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0)
489 		return 0;
490 	if ((vaddr_t)retaddr >= VM_MAXUSER_ADDRESS)
491 		return 0;
492 	tf->tf_rip = retaddr;
493 	tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr];
494 	tf->tf_rsp += 8;	/* "pop" the return address */
495 
496 #if 0
497 	printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n",
498 	    (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr,
499 	    vsyscallnr, (int)tf->tf_rax);
500 #endif
501 
502 	(*l->l_proc->p_md.md_syscall)(tf);
503 
504 	return 1;
505 }
506 
507 static void
508 linux_buildcontext(struct lwp *l, void *catcher, void *f)
509 {
510 	struct trapframe *tf = l->l_md.md_regs;
511 
512 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
513 	tf->tf_rip = (u_int64_t)catcher;
514 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
515 	tf->tf_rflags &= ~PSL_CLEARSIG;
516 	tf->tf_rsp = (u_int64_t)f;
517 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
518 }
519