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 #define __ELF_WORD_SIZE 64
36 #include <sys/param.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/kernel.h>
43 #include <sys/module.h>
44 #include <sys/mutex.h>
46 #include <sys/stddef.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
53 #include <vm/vm_param.h>
55 #include <machine/md_var.h>
56 #include <machine/trap.h>
58 #include <x86/linux/linux_x86.h>
59 #include <amd64/linux/linux.h>
60 #include <amd64/linux/linux_proto.h>
61 #include <compat/linux/linux_elf.h>
62 #include <compat/linux/linux_emul.h>
63 #include <compat/linux/linux_fork.h>
64 #include <compat/linux/linux_ioctl.h>
65 #include <compat/linux/linux_mib.h>
66 #include <compat/linux/linux_misc.h>
67 #include <compat/linux/linux_signal.h>
68 #include <compat/linux/linux_sysproto.h>
69 #include <compat/linux/linux_util.h>
70 #include <compat/linux/linux_vdso.h>
72 #include <x86/linux/linux_x86_sigframe.h>
74 _Static_assert(sizeof(struct l_fpstate) ==
75 sizeof(__typeof(((mcontext_t *)0)->mc_fpstate)),
76 "fxsave area size incorrect");
78 MODULE_VERSION(linux64, 1);
80 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2
81 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \
83 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE)
85 * PAGE_SIZE - the size
86 * of the native SHAREDPAGE
88 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48
89 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \
90 sizeof(struct ps_strings))
92 static int linux_szsigcode;
93 static vm_object_t linux_vdso_obj;
94 static char *linux_vdso_mapping;
95 extern char _binary_linux_vdso_so_o_start;
96 extern char _binary_linux_vdso_so_o_end;
97 static vm_offset_t linux_vdso_base;
99 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
100 extern const char *linux_syscallnames[];
102 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
104 static void linux_vdso_install(const void *param);
105 static void linux_vdso_deinstall(const void *param);
106 static void linux_vdso_reloc(char *mapping, Elf_Addr offset);
107 static void linux_set_syscall_retval(struct thread *td, int error);
108 static int linux_fetch_syscall_args(struct thread *td);
109 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
111 static void linux_exec_sysvec_init(void *param);
112 static int linux_on_exec_vmspace(struct proc *p,
113 struct image_params *imgp);
114 static void linux_set_fork_retval(struct thread *td);
115 static int linux_vsyscall(struct thread *td);
117 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
118 LINUX_VDSO_SYM_CHAR(linux_platform);
119 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
120 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
121 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
124 * According to the Intel x86 ISA 64-bit syscall
125 * saves %rip to %rcx and rflags to %r11. Registers on syscall entry:
126 * %rax system call number
127 * %rcx return address
136 * Then FreeBSD fast_syscall() move registers:
137 * %rcx -> trapframe.tf_rip
138 * %r10 -> trapframe.tf_rcx
141 linux_fetch_syscall_args(struct thread *td)
144 struct trapframe *frame;
145 struct syscall_args *sa;
148 frame = td->td_frame;
151 sa->args[0] = frame->tf_rdi;
152 sa->args[1] = frame->tf_rsi;
153 sa->args[2] = frame->tf_rdx;
154 sa->args[3] = frame->tf_rcx;
155 sa->args[4] = frame->tf_r8;
156 sa->args[5] = frame->tf_r9;
157 sa->code = frame->tf_rax;
158 sa->original_code = sa->code;
160 if (sa->code >= p->p_sysent->sv_size)
162 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
164 sa->callp = &p->p_sysent->sv_table[sa->code];
166 /* Restore r10 earlier to avoid doing this multiply times. */
167 frame->tf_r10 = frame->tf_rcx;
168 /* Restore %rcx for machine context. */
169 frame->tf_rcx = frame->tf_rip;
171 td->td_retval[0] = 0;
176 linux_set_syscall_retval(struct thread *td, int error)
178 struct trapframe *frame;
180 frame = td->td_frame;
184 frame->tf_rax = td->td_retval[0];
189 * Reconstruct pc, we know that 'syscall' is 2 bytes,
190 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
191 * We saved this in tf_err.
194 frame->tf_rip -= frame->tf_err;
201 frame->tf_rax = bsd_to_linux_errno(error);
206 * Differently from FreeBSD native ABI, on Linux only %rcx
207 * and %r11 values are not preserved across the syscall.
208 * Require full context restore to get all registers except
209 * those two restored at return to usermode.
211 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
215 linux_set_fork_retval(struct thread *td)
217 struct trapframe *frame = td->td_frame;
223 linux64_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
226 AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
227 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, cpu_feature);
228 AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, linux_x86_elf_hwcap2());
229 AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
233 * Reset registers to default values on exec.
236 linux_exec_setregs(struct thread *td, struct image_params *imgp,
239 struct trapframe *regs;
241 register_t saved_rflags;
246 if (td->td_proc->p_md.md_ldt != NULL)
251 clear_pcb_flags(pcb, PCB_32BIT);
252 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
253 set_pcb_flags(pcb, PCB_FULL_IRET);
255 saved_rflags = regs->tf_rflags & PSL_T;
256 bzero((char *)regs, sizeof(struct trapframe));
257 regs->tf_rip = imgp->entry_addr;
258 regs->tf_rsp = stack;
259 regs->tf_rflags = PSL_USER | saved_rflags;
260 regs->tf_ss = _udatasel;
261 regs->tf_cs = _ucodesel;
262 regs->tf_ds = _udatasel;
263 regs->tf_es = _udatasel;
264 regs->tf_fs = _ufssel;
265 regs->tf_gs = _ugssel;
266 regs->tf_flags = TF_HASSEGS;
268 x86_clear_dbregs(pcb);
271 * Drop the FP state if we hold it, so that the process gets a
272 * clean FP state if it uses the FPU again.
278 linux_fxrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc)
280 struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0];
283 error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate));
286 bzero(&fp->sv_pad[0], sizeof(fp->sv_pad));
287 return (set_fpcontext(td, mcp, NULL, 0));
291 linux_xrstor(struct thread *td, mcontext_t *mcp, struct l_sigcontext *sc)
293 struct savefpu *fp = (struct savefpu *)&mcp->mc_fpstate[0];
300 mcp->mc_xfpustate_len = cpu_max_ext_state_size - sizeof(struct savefpu);
302 /* Legacy region of an xsave area. */
303 error = copyin(PTRIN(sc->sc_fpstate), fp, sizeof(mcp->mc_fpstate));
306 bzero(&fp->sv_pad[0], sizeof(fp->sv_pad));
308 /* Extended region of an xsave area. */
309 sc->sc_fpstate += sizeof(mcp->mc_fpstate);
310 xfpustate = (char *)fpu_save_area_alloc();
311 error = copyin(PTRIN(sc->sc_fpstate), xfpustate, mcp->mc_xfpustate_len);
313 fpu_save_area_free((struct savefpu *)xfpustate);
314 uprintf("pid %d (%s): linux xrstor failed\n", p->p_pid,
319 /* Linux specific end of xsave area marker. */
320 sc->sc_fpstate += mcp->mc_xfpustate_len;
321 error = copyin(PTRIN(sc->sc_fpstate), &magic2, LINUX_FP_XSTATE_MAGIC2_SIZE);
322 if (error != 0 || magic2 != LINUX_FP_XSTATE_MAGIC2) {
323 fpu_save_area_free((struct savefpu *)xfpustate);
324 uprintf("pid %d (%s): sigreturn magic2 0x%x error %d\n",
325 p->p_pid, td->td_name, magic2, error);
329 error = set_fpcontext(td, mcp, xfpustate, mcp->mc_xfpustate_len);
330 fpu_save_area_free((struct savefpu *)xfpustate);
332 uprintf("pid %d (%s): sigreturn set_fpcontext error %d\n",
333 p->p_pid, td->td_name, error);
339 linux_copyin_fpstate(struct thread *td, struct l_ucontext *uc)
343 bzero(&mc, sizeof(mc));
344 mc.mc_ownedfp = _MC_FPOWNED_FPU;
345 mc.mc_fpformat = _MC_FPFMT_XMM;
347 if ((uc->uc_flags & LINUX_UC_FP_XSTATE) != 0)
348 return (linux_xrstor(td, &mc, &uc->uc_mcontext));
350 return (linux_fxrstor(td, &mc, &uc->uc_mcontext));
354 * Copied from amd64/amd64/machdep.c
357 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
360 struct l_rt_sigframe sf;
361 struct l_sigcontext *context;
362 struct trapframe *regs;
363 unsigned long rflags;
369 error = copyin((void *)regs->tf_rbx, &sf, sizeof(sf));
374 context = &sf.sf_uc.uc_mcontext;
375 rflags = context->sc_rflags;
378 * Don't allow users to change privileged or reserved flags.
381 * XXX do allow users to change the privileged flag PSL_RF.
382 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
383 * should sometimes set it there too. tf_rflags is kept in
384 * the signal context during signal handling and there is no
385 * other place to remember it, so the PSL_RF bit may be
386 * corrupted by the signal handler without us knowing.
387 * Corruption of the PSL_RF bit at worst causes one more or
388 * one less debugger trap, so allowing it is fairly harmless.
390 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
391 uprintf("pid %d comm %s linux mangled rflags %#lx\n",
392 p->p_pid, p->p_comm, rflags);
397 * Don't allow users to load a valid privileged %cs. Let the
398 * hardware check for invalid selectors, excess privilege in
399 * other selectors, invalid %eip's and invalid %esp's.
401 if (!CS_SECURE(context->sc_cs)) {
402 uprintf("pid %d comm %s linux mangled cs %#x\n",
403 p->p_pid, p->p_comm, context->sc_cs);
404 ksiginfo_init_trap(&ksi);
405 ksi.ksi_signo = SIGBUS;
406 ksi.ksi_code = BUS_OBJERR;
407 ksi.ksi_trapno = T_PROTFLT;
408 ksi.ksi_addr = (void *)regs->tf_rip;
409 trapsignal(td, &ksi);
413 linux_to_bsd_sigset(&sf.sf_uc.uc_sigmask, &bmask);
414 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
416 regs->tf_rdi = context->sc_rdi;
417 regs->tf_rsi = context->sc_rsi;
418 regs->tf_rdx = context->sc_rdx;
419 regs->tf_rbp = context->sc_rbp;
420 regs->tf_rbx = context->sc_rbx;
421 regs->tf_rcx = context->sc_rcx;
422 regs->tf_rax = context->sc_rax;
423 regs->tf_rip = context->sc_rip;
424 regs->tf_rsp = context->sc_rsp;
425 regs->tf_r8 = context->sc_r8;
426 regs->tf_r9 = context->sc_r9;
427 regs->tf_r10 = context->sc_r10;
428 regs->tf_r11 = context->sc_r11;
429 regs->tf_r12 = context->sc_r12;
430 regs->tf_r13 = context->sc_r13;
431 regs->tf_r14 = context->sc_r14;
432 regs->tf_r15 = context->sc_r15;
433 regs->tf_cs = context->sc_cs;
434 regs->tf_err = context->sc_err;
435 regs->tf_rflags = rflags;
437 error = linux_copyin_fpstate(td, &sf.sf_uc);
439 uprintf("pid %d comm %s linux can't restore fpu state %d\n",
440 p->p_pid, p->p_comm, error);
444 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
445 return (EJUSTRETURN);
449 linux_fxsave(mcontext_t *mcp, void *ufp)
451 struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0];
453 bzero(&fx->reserved2[0], sizeof(fx->reserved2));
454 return (copyout(fx, ufp, sizeof(*fx)));
458 linux_xsave(mcontext_t *mcp, char *xfpusave, char *ufp)
460 struct l_fpstate *fx = (struct l_fpstate *)&mcp->mc_fpstate[0];
464 /* Legacy region of an xsave area. */
465 fx->sw_reserved.magic1 = LINUX_FP_XSTATE_MAGIC1;
466 fx->sw_reserved.xstate_size = mcp->mc_xfpustate_len + sizeof(*fx);
467 fx->sw_reserved.extended_size = fx->sw_reserved.xstate_size +
468 LINUX_FP_XSTATE_MAGIC2_SIZE;
469 fx->sw_reserved.xfeatures = xsave_mask;
471 error = copyout(fx, ufp, sizeof(*fx));
476 /* Extended region of an xsave area. */
477 error = copyout(xfpusave, ufp, mcp->mc_xfpustate_len);
481 /* Linux specific end of xsave area marker. */
482 ufp += mcp->mc_xfpustate_len;
483 magic2 = LINUX_FP_XSTATE_MAGIC2;
484 return (copyout(&magic2, ufp, LINUX_FP_XSTATE_MAGIC2_SIZE));
488 linux_copyout_fpstate(struct thread *td, struct l_ucontext *uc, char **sp)
495 get_fpcontext(td, &mc, &xfpusave, &xfpusave_len);
496 KASSERT(mc.mc_fpformat != _MC_FPFMT_NODEV, ("fpu not present"));
498 /* Room for fxsave area. */
499 ufp -= sizeof(struct l_fpstate);
500 if (xfpusave != NULL) {
501 /* Room for xsave area. */
502 ufp -= (xfpusave_len + LINUX_FP_XSTATE_MAGIC2_SIZE);
503 uc->uc_flags |= LINUX_UC_FP_XSTATE;
505 *sp = ufp = (char *)((unsigned long)ufp & ~0x3Ful);
507 if (xfpusave != NULL)
508 return (linux_xsave(&mc, xfpusave, ufp));
510 return (linux_fxsave(&mc, ufp));
514 * copied from amd64/amd64/machdep.c
516 * Send an interrupt to process.
519 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
521 struct l_rt_sigframe sf, *sfp;
526 struct trapframe *regs;
528 int oonstack, issiginfo;
532 PROC_LOCK_ASSERT(p, MA_OWNED);
533 sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
535 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig);
536 code = ksi->ksi_code;
537 mtx_assert(&psp->ps_mtx, MA_OWNED);
539 oonstack = sigonstack(regs->tf_rsp);
541 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
542 catcher, sig, mask, code);
544 bzero(&sf, sizeof(sf));
545 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
546 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
547 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
548 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
550 /* Allocate space for the signal handler context. */
551 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
552 SIGISMEMBER(psp->ps_sigonstack, sig)) {
553 sp = (char *)td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
555 sp = (char *)regs->tf_rsp - 128;
557 mtx_unlock(&psp->ps_mtx);
560 if (linux_copyout_fpstate(td, &sf.sf_uc, &sp) != 0) {
561 uprintf("pid %d comm %s linux can't save fpu state, killing\n",
562 p->p_pid, p->p_comm);
566 sf.sf_uc.uc_mcontext.sc_fpstate = (register_t)sp;
568 /* Make room, keeping the stack aligned. */
569 sp -= sizeof(struct l_rt_sigframe);
570 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
572 /* Save user context. */
573 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask);
574 sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask;
575 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi;
576 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi;
577 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx;
578 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp;
579 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx;
580 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx;
581 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax;
582 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip;
583 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp;
584 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8;
585 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9;
586 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10;
587 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11;
588 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12;
589 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13;
590 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14;
591 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15;
592 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs;
593 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags;
594 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err;
595 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
596 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
598 /* Translate the signal. */
599 sig = bsd_to_linux_signal(sig);
600 /* Fill in POSIX parts. */
601 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig);
603 /* Copy the sigframe out to the user's stack. */
604 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
605 uprintf("pid %d comm %s has trashed its stack, killing\n",
606 p->p_pid, p->p_comm);
612 /* Build the argument list for the signal handler. */
613 regs->tf_rdi = sig; /* arg 1 in %rdi */
616 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
617 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */
622 regs->tf_rcx = (register_t)catcher;
623 regs->tf_rsp = (long)sfp;
624 regs->tf_rip = linux_rt_sigcode;
625 regs->tf_rflags &= ~(PSL_T | PSL_D);
626 regs->tf_cs = _ucodesel;
627 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
629 mtx_lock(&psp->ps_mtx);
632 #define LINUX_VSYSCALL_START (-10UL << 20)
633 #define LINUX_VSYSCALL_SZ 1024
635 const unsigned long linux_vsyscall_vector[] = {
636 LINUX_SYS_gettimeofday,
637 LINUX_SYS_linux_time,
638 LINUX_SYS_linux_getcpu,
642 linux_vsyscall(struct thread *td)
644 struct trapframe *frame;
649 frame = td->td_frame;
651 /* Check %rip for vsyscall area. */
652 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
654 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
656 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
657 if (code >= nitems(linux_vsyscall_vector))
661 * vsyscall called as callq *(%rax), so we must
662 * use return address from %rsp and also fixup %rsp.
664 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
668 frame->tf_rip = retqaddr;
669 frame->tf_rax = linux_vsyscall_vector[code];
672 traced = (frame->tf_flags & PSL_T);
674 amd64_syscall(td, traced);
679 struct sysentvec elf_linux_sysvec = {
680 .sv_size = LINUX_SYS_MAXSYSCALL,
681 .sv_table = linux_sysent,
682 .sv_fixup = __elfN(freebsd_fixup),
683 .sv_sendsig = linux_rt_sendsig,
684 .sv_sigcode = &_binary_linux_vdso_so_o_start,
685 .sv_szsigcode = &linux_szsigcode,
686 .sv_name = "Linux ELF64",
687 .sv_coredump = elf64_coredump,
688 .sv_elf_core_osabi = ELFOSABI_NONE,
689 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
690 .sv_elf_core_prepare_notes = linux64_prepare_notes,
691 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
692 .sv_minuser = VM_MIN_ADDRESS,
693 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
694 .sv_usrstack = LINUX_USRSTACK_LA48,
695 .sv_psstrings = LINUX_PS_STRINGS_LA48,
696 .sv_psstringssz = sizeof(struct ps_strings),
697 .sv_stackprot = VM_PROT_ALL,
698 .sv_copyout_auxargs = __linuxN(copyout_auxargs),
699 .sv_copyout_strings = __linuxN(copyout_strings),
700 .sv_setregs = linux_exec_setregs,
703 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
704 SV_SIG_WAITNDQ | SV_TIMEKEEP,
705 .sv_set_syscall_retval = linux_set_syscall_retval,
706 .sv_fetch_syscall_args = linux_fetch_syscall_args,
707 .sv_syscallnames = linux_syscallnames,
708 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48,
709 .sv_shared_page_len = PAGE_SIZE,
710 .sv_schedtail = linux_schedtail,
711 .sv_thread_detach = linux_thread_detach,
712 .sv_trap = linux_vsyscall,
715 .sv_onexec = linux_on_exec_vmspace,
716 .sv_onexit = linux_on_exit,
717 .sv_ontdexit = linux_thread_dtor,
718 .sv_setid_allowed = &linux_setid_allowed_query,
719 .sv_set_fork_retval = linux_set_fork_retval,
723 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
727 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
728 LINUX_VDSOPAGE_SIZE, imgp);
730 error = linux_on_exec(p, imgp);
735 * linux_vdso_install() and linux_exec_sysvec_init() must be called
736 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
739 linux_exec_sysvec_init(void *param)
741 l_uintptr_t *ktimekeep_base, *ktsc_selector;
742 struct sysentvec *sv;
746 amd64_lower_shared_page(sv);
747 /* Fill timekeep_base */
748 exec_sysvec_init(sv);
750 tkoff = kern_timekeep_base - linux_vdso_base;
751 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
752 *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
754 tkoff = kern_tsc_selector - linux_vdso_base;
755 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
756 *ktsc_selector = linux_vdso_tsc_selector_idx();
758 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector);
760 tkoff = kern_cpu_selector - linux_vdso_base;
761 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
762 *ktsc_selector = linux_vdso_cpu_selector_idx();
764 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector);
766 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
767 linux_exec_sysvec_init, &elf_linux_sysvec);
770 linux_vdso_install(const void *param)
772 char *vdso_start = &_binary_linux_vdso_so_o_start;
773 char *vdso_end = &_binary_linux_vdso_so_o_end;
775 linux_szsigcode = vdso_end - vdso_start;
776 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
778 linux_vdso_base = LINUX_VDSOPAGE_LA48;
779 if (hw_lower_amd64_sharedpage != 0)
780 linux_vdso_base -= PAGE_SIZE;
782 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
784 linux_vdso_obj = __elfN(linux_shared_page_init)
785 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
786 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
788 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
790 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
791 linux_vdso_install, NULL);
794 linux_vdso_deinstall(const void *param)
797 __elfN(linux_shared_page_fini)(linux_vdso_obj,
798 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
800 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
801 linux_vdso_deinstall, NULL);
804 linux_vdso_reloc(char *mapping, Elf_Addr offset)
806 const Elf_Ehdr *ehdr;
807 const Elf_Shdr *shdr;
808 Elf64_Addr *where, val;
809 Elf_Size rtype, symidx;
810 const Elf_Rela *rela;
811 Elf_Addr addr, addend;
818 ehdr = (const Elf_Ehdr *)mapping;
819 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
820 for (i = 0; i < ehdr->e_shnum; i++)
822 switch (shdr[i].sh_type) {
824 printf("Linux x86_64 vDSO: unexpected Rel section\n");
827 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
828 relacnt = shdr[i].sh_size / sizeof(*rela);
832 for (j = 0; j < relacnt; j++, rela++) {
833 where = (Elf_Addr *)(mapping + rela->r_offset);
834 addend = rela->r_addend;
835 rtype = ELF_R_TYPE(rela->r_info);
836 symidx = ELF_R_SYM(rela->r_info);
839 case R_X86_64_NONE: /* none */
842 case R_X86_64_RELATIVE: /* B + A */
843 addr = (Elf_Addr)(offset + addend);
848 case R_X86_64_IRELATIVE:
849 printf("Linux x86_64 vDSO: unexpected ifunc relocation, "
850 "symbol index %ld\n", symidx);
853 printf("Linux x86_64 vDSO: unexpected relocation type %ld, "
854 "symbol index %ld\n", rtype, symidx);
859 static Elf_Brandnote linux64_brandnote = {
860 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
863 .vendor = GNU_ABI_VENDOR,
864 .flags = BN_TRANSLATE_OSREL,
865 .trans_osrel = linux_trans_osrel
868 static Elf64_Brandinfo linux_glibc2brand = {
869 .brand = ELFOSABI_LINUX,
870 .machine = EM_X86_64,
871 .compat_3_brand = "Linux",
872 .interp_path = "/lib64/ld-linux-x86-64.so.2",
873 .sysvec = &elf_linux_sysvec,
874 .interp_newpath = NULL,
875 .brand_note = &linux64_brandnote,
876 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
879 static Elf64_Brandinfo linux_glibc2brandshort = {
880 .brand = ELFOSABI_LINUX,
881 .machine = EM_X86_64,
882 .compat_3_brand = "Linux",
883 .interp_path = "/lib64/ld-linux.so.2",
884 .sysvec = &elf_linux_sysvec,
885 .interp_newpath = NULL,
886 .brand_note = &linux64_brandnote,
887 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
890 static Elf64_Brandinfo linux_muslbrand = {
891 .brand = ELFOSABI_LINUX,
892 .machine = EM_X86_64,
893 .compat_3_brand = "Linux",
894 .interp_path = "/lib/ld-musl-x86_64.so.1",
895 .sysvec = &elf_linux_sysvec,
896 .interp_newpath = NULL,
897 .brand_note = &linux64_brandnote,
898 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
899 LINUX_BI_FUTEX_REQUEUE
902 Elf64_Brandinfo *linux_brandlist[] = {
904 &linux_glibc2brandshort,
910 linux64_elf_modevent(module_t mod, int type, void *data)
912 Elf64_Brandinfo **brandinfo;
914 struct linux_ioctl_handler **lihp;
920 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
922 if (elf64_insert_brand_entry(*brandinfo) < 0)
925 SET_FOREACH(lihp, linux_ioctl_handler_set)
926 linux_ioctl_register_handler(*lihp);
927 stclohz = (stathz ? stathz : hz);
929 printf("Linux x86-64 ELF exec handler installed\n");
931 printf("cannot insert Linux x86-64 ELF brand handler\n");
934 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
936 if (elf64_brand_inuse(*brandinfo))
939 for (brandinfo = &linux_brandlist[0];
940 *brandinfo != NULL; ++brandinfo)
941 if (elf64_remove_brand_entry(*brandinfo) < 0)
945 SET_FOREACH(lihp, linux_ioctl_handler_set)
946 linux_ioctl_unregister_handler(*lihp);
948 printf("Linux x86_64 ELF exec handler removed\n");
950 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n");
958 static moduledata_t linux64_elf_mod = {
960 linux64_elf_modevent,
964 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
965 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
966 FEATURE(linux64, "Linux 64bit support");