2 * Copyright (c) 2004 Tim J. Robbins
3 * Copyright (c) 2003 Peter Wemm
4 * Copyright (c) 2002 Doug Rabson
5 * Copyright (c) 1998-1999 Andrew Gallatin
6 * Copyright (c) 1994-1996 Søren Schmidt
8 * Copyright (c) 2013, 2021 Dmitry Chagin <dchagin@FreeBSD.org>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 #define __ELF_WORD_SIZE 64
37 #include <sys/param.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/module.h>
46 #include <sys/mutex.h>
48 #include <sys/stddef.h>
49 #include <sys/syscallsubr.h>
50 #include <sys/sysctl.h>
51 #include <sys/sysent.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_page.h>
58 #include <machine/cpu.h>
59 #include <machine/md_var.h>
60 #include <machine/pcb.h>
61 #include <machine/specialreg.h>
62 #include <machine/trap.h>
64 #include <x86/linux/linux_x86.h>
65 #include <amd64/linux/linux.h>
66 #include <amd64/linux/linux_proto.h>
67 #include <compat/linux/linux_elf.h>
68 #include <compat/linux/linux_emul.h>
69 #include <compat/linux/linux_fork.h>
70 #include <compat/linux/linux_ioctl.h>
71 #include <compat/linux/linux_mib.h>
72 #include <compat/linux/linux_misc.h>
73 #include <compat/linux/linux_signal.h>
74 #include <compat/linux/linux_sysproto.h>
75 #include <compat/linux/linux_util.h>
76 #include <compat/linux/linux_vdso.h>
78 #include <x86/linux/linux_x86_sigframe.h>
80 _Static_assert(sizeof(struct l_fpstate) ==
81 sizeof(__typeof(((mcontext_t *)0)->mc_fpstate)),
82 "fxsave area size incorrect");
84 MODULE_VERSION(linux64, 1);
86 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2
87 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \
89 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE)
91 * PAGE_SIZE - the size
92 * of the native SHAREDPAGE
94 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48
95 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \
96 sizeof(struct ps_strings))
98 static int linux_szsigcode;
99 static vm_object_t linux_vdso_obj;
100 static char *linux_vdso_mapping;
101 extern char _binary_linux_vdso_so_o_start;
102 extern char _binary_linux_vdso_so_o_end;
103 static vm_offset_t linux_vdso_base;
105 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
106 extern const char *linux_syscallnames[];
108 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
110 static void linux_vdso_install(const void *param);
111 static void linux_vdso_deinstall(const void *param);
112 static void linux_vdso_reloc(char *mapping, Elf_Addr offset);
113 static void linux_set_syscall_retval(struct thread *td, int error);
114 static int linux_fetch_syscall_args(struct thread *td);
115 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
117 static void linux_exec_sysvec_init(void *param);
118 static int linux_on_exec_vmspace(struct proc *p,
119 struct image_params *imgp);
120 static void linux_set_fork_retval(struct thread *td);
121 static int linux_vsyscall(struct thread *td);
123 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
124 LINUX_VDSO_SYM_CHAR(linux_platform);
125 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
126 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
127 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
130 * According to the Intel x86 ISA 64-bit syscall
131 * saves %rip to %rcx and rflags to %r11. Registers on syscall entry:
132 * %rax system call number
133 * %rcx return address
142 * Then FreeBSD fast_syscall() move registers:
143 * %rcx -> trapframe.tf_rip
144 * %r10 -> trapframe.tf_rcx
147 linux_fetch_syscall_args(struct thread *td)
150 struct trapframe *frame;
151 struct syscall_args *sa;
154 frame = td->td_frame;
157 sa->args[0] = frame->tf_rdi;
158 sa->args[1] = frame->tf_rsi;
159 sa->args[2] = frame->tf_rdx;
160 sa->args[3] = frame->tf_rcx;
161 sa->args[4] = frame->tf_r8;
162 sa->args[5] = frame->tf_r9;
163 sa->code = frame->tf_rax;
164 sa->original_code = sa->code;
166 if (sa->code >= p->p_sysent->sv_size)
168 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
170 sa->callp = &p->p_sysent->sv_table[sa->code];
172 /* Restore r10 earlier to avoid doing this multiply times. */
173 frame->tf_r10 = frame->tf_rcx;
174 /* Restore %rcx for machine context. */
175 frame->tf_rcx = frame->tf_rip;
177 td->td_retval[0] = 0;
182 linux_set_syscall_retval(struct thread *td, int error)
184 struct trapframe *frame;
186 frame = td->td_frame;
190 frame->tf_rax = td->td_retval[0];
195 * Reconstruct pc, we know that 'syscall' is 2 bytes,
196 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
197 * We saved this in tf_err.
200 frame->tf_rip -= frame->tf_err;
207 frame->tf_rax = bsd_to_linux_errno(error);
212 * Differently from FreeBSD native ABI, on Linux only %rcx
213 * and %r11 values are not preserved across the syscall.
214 * Require full context restore to get all registers except
215 * those two restored at return to usermode.
217 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
221 linux_set_fork_retval(struct thread *td)
223 struct trapframe *frame = td->td_frame;
229 linux64_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
232 AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
233 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature);
234 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2());
235 AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
239 * Reset registers to default values on exec.
242 linux_exec_setregs(struct thread *td, struct image_params *imgp,
245 struct trapframe *regs;
247 register_t saved_rflags;
252 if (td->td_proc->p_md.md_ldt != NULL)
257 clear_pcb_flags(pcb, PCB_32BIT);
258 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
259 set_pcb_flags(pcb, PCB_FULL_IRET);
261 saved_rflags = regs->tf_rflags & PSL_T;
262 bzero((char *)regs, sizeof(struct trapframe));
263 regs->tf_rip = imgp->entry_addr;
264 regs->tf_rsp = stack;
265 regs->tf_rflags = PSL_USER | saved_rflags;
266 regs->tf_ss = _udatasel;
267 regs->tf_cs = _ucodesel;
268 regs->tf_ds = _udatasel;
269 regs->tf_es = _udatasel;
270 regs->tf_fs = _ufssel;
271 regs->tf_gs = _ugssel;
272 regs->tf_flags = TF_HASSEGS;
274 x86_clear_dbregs(pcb);
277 * Drop the FP state if we hold it, so that the process gets a
278 * clean FP state if it uses the FPU again.
284 linux_fxrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc)
286 struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0];
289 error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate));
292 bzero(&fp->sv_pad[0], sizeof(fp->sv_pad));
293 return (set_fpcontext(td, mcp, NULL, 0));
297 linux_xrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc)
299 struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0];
306 mcp->mc_xfpustate_len = cpu_max_ext_state_size - sizeof(struct savefpu);
308 /* Legacy region of an xsave area. */
309 error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate));
312 bzero(&fp->sv_pad[0], sizeof(fp->sv_pad));
314 /* Extended region of an xsave area. */
315 sc->sc_fpstate += sizeof(mcp->mc_fpstate);
316 xfpustate = (char *)fpu_save_area_alloc();
317 error = copyin(PTRIN(sc->sc_fpstate), xfpustate, mcp->mc_xfpustate_len);
319 fpu_save_area_free((struct savefpu *)xfpustate);
320 uprintf("pid %d (%s): linux xrstor failed\n", p->p_pid,
325 /* Linux specific end of xsave area marker. */
326 sc->sc_fpstate += mcp->mc_xfpustate_len;
327 error = copyin(PTRIN(sc->sc_fpstate), &magic2, LINUX_FP_XSTATE_MAGIC2_SIZE);
328 if (error != 0 || magic2 != LINUX_FP_XSTATE_MAGIC2) {
329 fpu_save_area_free((struct savefpu *)xfpustate);
330 uprintf("pid %d (%s): sigreturn magic2 0x%x error %d\n",
331 p->p_pid, td->td_name, magic2, error);
335 error = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len);
336 fpu_save_area_free((struct savefpu *)xfpustate);
338 uprintf("pid %d (%s): sigreturn set_fpcontext error %d\n",
339 p->p_pid, td->td_name, error);
345 linux_copyin_fpstate(struct thread *td, struct l_ucontext *uc)
349 bzero(&mc, sizeof(mc));
350 mc.mc_ownedfp = _MC_FPOWNED_FPU;
351 mc.mc_fpformat = _MC_FPFMT_XMM;
353 if ((uc->uc_flags & LINUX_UC_FP_XSTATE) != 0)
354 return (linux_xrstor(td, &mc, &uc->uc_mcontext));
356 return (linux_fxrstor(td, &mc, &uc->uc_mcontext));
360 * Copied from amd64/amd64/machdep.c
363 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
366 struct l_rt_sigframe sf;
367 struct l_sigcontext *context;
368 struct trapframe *regs;
369 unsigned long rflags;
375 error = copyin((void *)regs->tf_rbx, &sf, sizeof(sf));
380 context = &sf.sf_uc.uc_mcontext;
381 rflags = context->sc_rflags;
384 * Don't allow users to change privileged or reserved flags.
387 * XXX do allow users to change the privileged flag PSL_RF.
388 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
389 * should sometimes set it there too. tf_rflags is kept in
390 * the signal context during signal handling and there is no
391 * other place to remember it, so the PSL_RF bit may be
392 * corrupted by the signal handler without us knowing.
393 * Corruption of the PSL_RF bit at worst causes one more or
394 * one less debugger trap, so allowing it is fairly harmless.
396 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
397 uprintf("pid %d comm %s linux mangled rflags %#lx\n",
398 p->p_pid, p->p_comm, rflags);
403 * Don't allow users to load a valid privileged %cs. Let the
404 * hardware check for invalid selectors, excess privilege in
405 * other selectors, invalid %eip's and invalid %esp's.
407 if (!CS_SECURE(context->sc_cs)) {
408 uprintf("pid %d comm %s linux mangled cs %#x\n",
409 p->p_pid, p->p_comm, context->sc_cs);
410 ksiginfo_init_trap(&ksi);
411 ksi.ksi_signo = SIGBUS;
412 ksi.ksi_code = BUS_OBJERR;
413 ksi.ksi_trapno = T_PROTFLT;
414 ksi.ksi_addr = (void *)regs->tf_rip;
415 trapsignal(td, &ksi);
419 linux_to_bsd_sigset(&sf.sf_uc.uc_sigmask, &bmask);
420 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
422 regs->tf_rdi = context->sc_rdi;
423 regs->tf_rsi = context->sc_rsi;
424 regs->tf_rdx = context->sc_rdx;
425 regs->tf_rbp = context->sc_rbp;
426 regs->tf_rbx = context->sc_rbx;
427 regs->tf_rcx = context->sc_rcx;
428 regs->tf_rax = context->sc_rax;
429 regs->tf_rip = context->sc_rip;
430 regs->tf_rsp = context->sc_rsp;
431 regs->tf_r8 = context->sc_r8;
432 regs->tf_r9 = context->sc_r9;
433 regs->tf_r10 = context->sc_r10;
434 regs->tf_r11 = context->sc_r11;
435 regs->tf_r12 = context->sc_r12;
436 regs->tf_r13 = context->sc_r13;
437 regs->tf_r14 = context->sc_r14;
438 regs->tf_r15 = context->sc_r15;
439 regs->tf_cs = context->sc_cs;
440 regs->tf_err = context->sc_err;
441 regs->tf_rflags = rflags;
443 error = linux_copyin_fpstate(td, &sf.sf_uc);
445 uprintf("pid %d comm %s linux can't restore fpu state %d\n",
446 p->p_pid, p->p_comm, error);
450 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
451 return (EJUSTRETURN);
455 linux_fxsave(mcontext_t *mcp, void *ufp)
457 struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0];
459 bzero(&fx->reserved2[0], sizeof(fx->reserved2));
460 return (copyout(fx, ufp, sizeof(*fx)));
464 linux_xsave(mcontext_t *mcp, char *xfpusave, char *ufp)
466 struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0];
470 /* Legacy region of an xsave area. */
471 fx->sw_reserved.magic1 = LINUX_FP_XSTATE_MAGIC1;
472 fx->sw_reserved.xstate_size = mcp->mc_xfpustate_len + sizeof(*fx);
473 fx->sw_reserved.extended_size = fx->sw_reserved.xstate_size +
474 LINUX_FP_XSTATE_MAGIC2_SIZE;
475 fx->sw_reserved.xfeatures = xsave_mask;
477 error = copyout(fx, ufp, sizeof(*fx));
482 /* Extended region of an xsave area. */
483 error = copyout(xfpusave, ufp, mcp->mc_xfpustate_len);
487 /* Linux specific end of xsave area marker. */
488 ufp += mcp->mc_xfpustate_len;
489 magic2 = LINUX_FP_XSTATE_MAGIC2;
490 return (copyout(&magic2, ufp, LINUX_FP_XSTATE_MAGIC2_SIZE));
494 linux_copyout_fpstate(struct thread *td, struct l_ucontext *uc, char **sp)
501 get_fpcontext(td, &mc, &xfpusave, &xfpusave_len);
502 KASSERT(mc.mc_fpformat != _MC_FPFMT_NODEV, ("fpu not present"));
504 /* Room for fxsave area. */
505 ufp -= sizeof(struct l_fpstate);
506 if (xfpusave != NULL) {
507 /* Room for xsave area. */
508 ufp -= (xfpusave_len + LINUX_FP_XSTATE_MAGIC2_SIZE);
509 uc->uc_flags |= LINUX_UC_FP_XSTATE;
511 *sp = ufp = (char *)((unsigned long)ufp & ~0x3Ful);
513 if (xfpusave != NULL)
514 return (linux_xsave(&mc, xfpusave, ufp));
516 return (linux_fxsave(&mc, ufp));
520 * copied from amd64/amd64/machdep.c
522 * Send an interrupt to process.
525 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
527 struct l_rt_sigframe sf, *sfp;
532 struct trapframe *regs;
534 int oonstack, issiginfo;
538 PROC_LOCK_ASSERT(p, MA_OWNED);
539 sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
541 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig);
542 code = ksi->ksi_code;
543 mtx_assert(&psp->ps_mtx, MA_OWNED);
545 oonstack = sigonstack(regs->tf_rsp);
547 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
548 catcher, sig, mask, code);
550 bzero(&sf, sizeof(sf));
551 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
552 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
553 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
554 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
556 /* Allocate space for the signal handler context. */
557 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
558 SIGISMEMBER(psp->ps_sigonstack, sig)) {
559 sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
561 sp = (char *)regs->tf_rsp - 128;
563 mtx_unlock(&psp->ps_mtx);
566 if (linux_copyout_fpstate(td, &sf.sf_uc, &sp) != 0) {
567 uprintf("pid %d comm %s linux can't save fpu state, killing\n",
568 p->p_pid, p->p_comm);
572 sf.sf_uc.uc_mcontext.sc_fpstate = (register_t)sp;
574 /* Make room, keeping the stack aligned. */
575 sp -= sizeof(struct l_rt_sigframe);
576 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
578 /* Save user context. */
579 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask);
580 sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask;
581 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi;
582 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi;
583 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx;
584 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp;
585 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx;
586 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx;
587 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax;
588 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip;
589 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp;
590 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8;
591 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9;
592 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10;
593 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11;
594 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12;
595 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13;
596 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14;
597 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15;
598 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs;
599 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags;
600 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err;
601 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
602 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
604 /* Translate the signal. */
605 sig = bsd_to_linux_signal(sig);
606 /* Fill in POSIX parts. */
607 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig);
609 /* Copy the sigframe out to the user's stack. */
610 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
611 uprintf("pid %d comm %s has trashed its stack, killing\n",
612 p->p_pid, p->p_comm);
618 /* Build the argument list for the signal handler. */
619 regs->tf_rdi = sig; /* arg 1 in %rdi */
622 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
623 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */
628 regs->tf_rcx = (register_t)catcher;
629 regs->tf_rsp = (long)sfp;
630 regs->tf_rip = linux_rt_sigcode;
631 regs->tf_rflags &= ~(PSL_T | PSL_D);
632 regs->tf_cs = _ucodesel;
633 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
635 mtx_lock(&psp->ps_mtx);
638 #define LINUX_VSYSCALL_START (-10UL << 20)
639 #define LINUX_VSYSCALL_SZ 1024
641 const unsigned long linux_vsyscall_vector[] = {
642 LINUX_SYS_gettimeofday,
643 LINUX_SYS_linux_time,
644 LINUX_SYS_linux_getcpu,
648 linux_vsyscall(struct thread *td)
650 struct trapframe *frame;
655 frame = td->td_frame;
657 /* Check %rip for vsyscall area. */
658 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
660 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
662 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
663 if (code >= nitems(linux_vsyscall_vector))
667 * vsyscall called as callq *(%rax), so we must
668 * use return address from %rsp and also fixup %rsp.
670 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
674 frame->tf_rip = retqaddr;
675 frame->tf_rax = linux_vsyscall_vector[code];
678 traced = (frame->tf_flags & PSL_T);
680 amd64_syscall(td, traced);
685 struct sysentvec elf_linux_sysvec = {
686 .sv_size = LINUX_SYS_MAXSYSCALL,
687 .sv_table = linux_sysent,
688 .sv_fixup = __elfN(freebsd_fixup),
689 .sv_sendsig = linux_rt_sendsig,
690 .sv_sigcode = &_binary_linux_vdso_so_o_start,
691 .sv_szsigcode = &linux_szsigcode,
692 .sv_name = "Linux ELF64",
693 .sv_coredump = elf64_coredump,
694 .sv_elf_core_osabi = ELFOSABI_NONE,
695 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
696 .sv_elf_core_prepare_notes = linux64_prepare_notes,
697 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
698 .sv_minuser = VM_MIN_ADDRESS,
699 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
700 .sv_usrstack = LINUX_USRSTACK_LA48,
701 .sv_psstrings = LINUX_PS_STRINGS_LA48,
702 .sv_psstringssz = sizeof(struct ps_strings),
703 .sv_stackprot = VM_PROT_ALL,
704 .sv_copyout_auxargs = __linuxN(copyout_auxargs),
705 .sv_copyout_strings = __linuxN(copyout_strings),
706 .sv_setregs = linux_exec_setregs,
709 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
710 SV_SIG_WAITNDQ | SV_TIMEKEEP,
711 .sv_set_syscall_retval = linux_set_syscall_retval,
712 .sv_fetch_syscall_args = linux_fetch_syscall_args,
713 .sv_syscallnames = linux_syscallnames,
714 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48,
715 .sv_shared_page_len = PAGE_SIZE,
716 .sv_schedtail = linux_schedtail,
717 .sv_thread_detach = linux_thread_detach,
718 .sv_trap = linux_vsyscall,
721 .sv_onexec = linux_on_exec_vmspace,
722 .sv_onexit = linux_on_exit,
723 .sv_ontdexit = linux_thread_dtor,
724 .sv_setid_allowed = &linux_setid_allowed_query,
725 .sv_set_fork_retval = linux_set_fork_retval,
729 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
733 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
734 LINUX_VDSOPAGE_SIZE, imgp);
736 error = linux_on_exec(p, imgp);
741 * linux_vdso_install() and linux_exec_sysvec_init() must be called
742 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
745 linux_exec_sysvec_init(void *param)
747 l_uintptr_t *ktimekeep_base, *ktsc_selector;
748 struct sysentvec *sv;
752 amd64_lower_shared_page(sv);
753 /* Fill timekeep_base */
754 exec_sysvec_init(sv);
756 tkoff = kern_timekeep_base - linux_vdso_base;
757 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
758 *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
760 tkoff = kern_tsc_selector - linux_vdso_base;
761 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
762 *ktsc_selector = linux_vdso_tsc_selector_idx();
764 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector);
766 tkoff = kern_cpu_selector - linux_vdso_base;
767 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
768 *ktsc_selector = linux_vdso_cpu_selector_idx();
770 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector);
772 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
773 linux_exec_sysvec_init, &elf_linux_sysvec);
776 linux_vdso_install(const void *param)
778 char *vdso_start = &_binary_linux_vdso_so_o_start;
779 char *vdso_end = &_binary_linux_vdso_so_o_end;
781 linux_szsigcode = vdso_end - vdso_start;
782 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
784 linux_vdso_base = LINUX_VDSOPAGE_LA48;
785 if (hw_lower_amd64_sharedpage != 0)
786 linux_vdso_base -= PAGE_SIZE;
788 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
790 linux_vdso_obj = __elfN(linux_shared_page_init)
791 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
792 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
794 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
796 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
797 linux_vdso_install, NULL);
800 linux_vdso_deinstall(const void *param)
803 __elfN(linux_shared_page_fini)(linux_vdso_obj,
804 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
806 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
807 linux_vdso_deinstall, NULL);
810 linux_vdso_reloc(char *mapping, Elf_Addr offset)
812 const Elf_Ehdr *ehdr;
813 const Elf_Shdr *shdr;
814 Elf64_Addr *where, val;
815 Elf_Size rtype, symidx;
816 const Elf_Rela *rela;
817 Elf_Addr addr, addend;
824 ehdr = (const Elf_Ehdr *)mapping;
825 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
826 for (i = 0; i < ehdr->e_shnum; i++)
828 switch (shdr[i].sh_type) {
830 printf("Linux x86_64 vDSO: unexpected Rel section\n");
833 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
834 relacnt = shdr[i].sh_size / sizeof(*rela);
838 for (j = 0; j < relacnt; j++, rela++) {
839 where = (Elf_Addr *)(mapping + rela->r_offset);
840 addend = rela->r_addend;
841 rtype = ELF_R_TYPE(rela->r_info);
842 symidx = ELF_R_SYM(rela->r_info);
845 case R_X86_64_NONE: /* none */
848 case R_X86_64_RELATIVE: /* B + A */
849 addr = (Elf_Addr)(offset + addend);
854 case R_X86_64_IRELATIVE:
855 printf("Linux x86_64 vDSO: unexpected ifunc relocation, "
856 "symbol index %ld\n", symidx);
859 printf("Linux x86_64 vDSO: unexpected relocation type %ld, "
860 "symbol index %ld\n", rtype, symidx);
865 static Elf_Brandnote linux64_brandnote = {
866 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
869 .vendor = GNU_ABI_VENDOR,
870 .flags = BN_TRANSLATE_OSREL,
871 .trans_osrel = linux_trans_osrel
874 static Elf64_Brandinfo linux_glibc2brand = {
875 .brand = ELFOSABI_LINUX,
876 .machine = EM_X86_64,
877 .compat_3_brand = "Linux",
878 .interp_path = "/lib64/ld-linux-x86-64.so.2",
879 .sysvec = &elf_linux_sysvec,
880 .interp_newpath = NULL,
881 .brand_note = &linux64_brandnote,
882 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
885 static Elf64_Brandinfo linux_glibc2brandshort = {
886 .brand = ELFOSABI_LINUX,
887 .machine = EM_X86_64,
888 .compat_3_brand = "Linux",
889 .interp_path = "/lib64/ld-linux.so.2",
890 .sysvec = &elf_linux_sysvec,
891 .interp_newpath = NULL,
892 .brand_note = &linux64_brandnote,
893 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
896 static Elf64_Brandinfo linux_muslbrand = {
897 .brand = ELFOSABI_LINUX,
898 .machine = EM_X86_64,
899 .compat_3_brand = "Linux",
900 .interp_path = "/lib/ld-musl-x86_64.so.1",
901 .sysvec = &elf_linux_sysvec,
902 .interp_newpath = NULL,
903 .brand_note = &linux64_brandnote,
904 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
905 LINUX_BI_FUTEX_REQUEUE
908 Elf64_Brandinfo *linux_brandlist[] = {
910 &linux_glibc2brandshort,
916 linux64_elf_modevent(module_t mod, int type, void *data)
918 Elf64_Brandinfo **brandinfo;
920 struct linux_ioctl_handler **lihp;
926 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
928 if (elf64_insert_brand_entry(*brandinfo) < 0)
931 SET_FOREACH(lihp, linux_ioctl_handler_set)
932 linux_ioctl_register_handler(*lihp);
933 stclohz = (stathz ? stathz : hz);
935 printf("Linux x86-64 ELF exec handler installed\n");
937 printf("cannot insert Linux x86-64 ELF brand handler\n");
940 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
942 if (elf64_brand_inuse(*brandinfo))
945 for (brandinfo = &linux_brandlist[0];
946 *brandinfo != NULL; ++brandinfo)
947 if (elf64_remove_brand_entry(*brandinfo) < 0)
951 SET_FOREACH(lihp, linux_ioctl_handler_set)
952 linux_ioctl_unregister_handler(*lihp);
954 printf("Linux x86_64 ELF exec handler removed\n");
956 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n");
964 static moduledata_t linux64_elf_mod = {
966 linux64_elf_modevent,
970 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
971 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
972 FEATURE(linux64, "Linux 64bit support");