Modify the trapframe sigcontext, ucontext, etc. Add %gs to the trapframe
[dragonfly.git] / sys / platform / vkernel / i386 / cpu_regs.c
CommitLineData
6f7b98e0
MD
1/*-
2 * Copyright (c) 1992 Terrence R. Lambert.
3 * Copyright (C) 1994, David Greenman
4 * Copyright (c) 1982, 1987, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * William Jolitz.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * from: @(#)machdep.c 7.4 (Berkeley) 6/3/91
39 * $FreeBSD: src/sys/i386/i386/machdep.c,v 1.385.2.30 2003/05/31 08:48:05 alc Exp $
4e7c41c5 40 * $DragonFly: src/sys/platform/vkernel/i386/cpu_regs.c,v 1.4 2007/01/08 03:33:43 dillon Exp $
6f7b98e0
MD
41 */
42
43#include "use_ether.h"
44#include "use_npx.h"
45#include "use_isa.h"
46#include "opt_atalk.h"
47#include "opt_compat.h"
48#include "opt_ddb.h"
49#include "opt_directio.h"
50#include "opt_inet.h"
51#include "opt_ipx.h"
52#include "opt_msgbuf.h"
53#include "opt_swap.h"
54
55#include <sys/param.h>
56#include <sys/systm.h>
57#include <sys/sysproto.h>
58#include <sys/signalvar.h>
59#include <sys/kernel.h>
60#include <sys/linker.h>
61#include <sys/malloc.h>
62#include <sys/proc.h>
63#include <sys/buf.h>
64#include <sys/reboot.h>
65#include <sys/mbuf.h>
66#include <sys/msgbuf.h>
67#include <sys/sysent.h>
68#include <sys/sysctl.h>
69#include <sys/vmmeter.h>
70#include <sys/bus.h>
71#include <sys/upcall.h>
72#include <sys/usched.h>
73#include <sys/reg.h>
74
75#include <vm/vm.h>
76#include <vm/vm_param.h>
77#include <sys/lock.h>
78#include <vm/vm_kern.h>
79#include <vm/vm_object.h>
80#include <vm/vm_page.h>
81#include <vm/vm_map.h>
82#include <vm/vm_pager.h>
83#include <vm/vm_extern.h>
84
85#include <sys/thread2.h>
86
87#include <sys/user.h>
88#include <sys/exec.h>
89#include <sys/cons.h>
90
91#include <ddb/ddb.h>
92
93#include <machine/cpu.h>
94#include <machine/clock.h>
95#include <machine/specialreg.h>
96#include <machine/md_var.h>
97#include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */
98#include <machine/globaldata.h> /* CPU_prvspace */
99#include <machine/smp.h>
100#ifdef PERFMON
101#include <machine/perfmon.h>
102#endif
103#include <machine/cputypes.h>
104
105#include <bus/isa/rtc.h>
106#include <machine/vm86.h>
107#include <sys/random.h>
108#include <sys/ptrace.h>
109#include <machine/sigframe.h>
110
111extern void dblfault_handler (void);
112
113#ifndef CPU_DISABLE_SSE
114static void set_fpregs_xmm (struct save87 *, struct savexmm *);
115static void fill_fpregs_xmm (struct savexmm *, struct save87 *);
116#endif /* CPU_DISABLE_SSE */
117#ifdef DIRECTIO
118extern void ffs_rawread_setup(void);
119#endif /* DIRECTIO */
120
121#ifdef SMP
122int64_t tsc_offsets[MAXCPU];
123#else
124int64_t tsc_offsets[1];
125#endif
126
127#if defined(SWTCH_OPTIM_STATS)
128extern int swtch_optim_stats;
129SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats,
130 CTLFLAG_RD, &swtch_optim_stats, 0, "");
131SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count,
132 CTLFLAG_RD, &tlb_flush_count, 0, "");
133#endif
134
6f7b98e0
MD
135static int
136sysctl_hw_physmem(SYSCTL_HANDLER_ARGS)
137{
71152ac6 138 int error = sysctl_handle_int(oidp, 0, ctob((int)Maxmem), req);
6f7b98e0
MD
139 return (error);
140}
141
142SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_INT|CTLFLAG_RD,
143 0, 0, sysctl_hw_physmem, "IU", "");
144
145static int
146sysctl_hw_usermem(SYSCTL_HANDLER_ARGS)
147{
148 int error = sysctl_handle_int(oidp, 0,
71152ac6 149 ctob((int)Maxmem - vmstats.v_wire_count), req);
6f7b98e0
MD
150 return (error);
151}
152
153SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD,
154 0, 0, sysctl_hw_usermem, "IU", "");
155
156#if 0
157
158static int
159sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS)
160{
161 int error;
162
163 /* Unwind the buffer, so that it's linear (possibly starting with
164 * some initial nulls).
165 */
166 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr,
167 msgbufp->msg_size-msgbufp->msg_bufr,req);
168 if(error) return(error);
169 if(msgbufp->msg_bufr>0) {
170 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr,
171 msgbufp->msg_bufr,req);
172 }
173 return(error);
174}
175
176SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD,
177 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer");
178
179static int msgbuf_clear;
180
181static int
182sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS)
183{
184 int error;
185 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
186 req);
187 if (!error && req->newptr) {
188 /* Clear the buffer and reset write pointer */
189 bzero(msgbufp->msg_ptr,msgbufp->msg_size);
190 msgbufp->msg_bufr=msgbufp->msg_bufx=0;
191 msgbuf_clear=0;
192 }
193 return (error);
194}
195
196SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW,
197 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I",
198 "Clear kernel message buffer");
199
200#endif
201
202/*
203 * Send an interrupt to process.
204 *
205 * Stack is set up to allow sigcode stored
206 * at top to call routine, followed by kcall
207 * to sigreturn routine below. After sigreturn
208 * resets the signal mask, the stack, and the
209 * frame pointer, it returns to the user
210 * specified pc, psl.
211 */
212void
213sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code)
214{
215 struct lwp *lp = curthread->td_lwp;
216 struct proc *p = lp->lwp_proc;
217 struct trapframe *regs;
218 struct sigacts *psp = p->p_sigacts;
219 struct sigframe sf, *sfp;
220 int oonstack;
221
222 regs = lp->lwp_md.md_regs;
223 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0;
224
225 /* save user context */
226 bzero(&sf, sizeof(struct sigframe));
227 sf.sf_uc.uc_sigmask = *mask;
228 sf.sf_uc.uc_stack = lp->lwp_sigstk;
229 sf.sf_uc.uc_mcontext.mc_onstack = oonstack;
4e7c41c5 230 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_gs, sizeof(struct trapframe));
6f7b98e0
MD
231
232 /* Allocate and validate space for the signal handler context. */
233 /* XXX lwp flags */
234 if ((p->p_flag & P_ALTSTACK) != 0 && !oonstack &&
235 SIGISMEMBER(psp->ps_sigonstack, sig)) {
236 sfp = (struct sigframe *)(lp->lwp_sigstk.ss_sp +
237 lp->lwp_sigstk.ss_size - sizeof(struct sigframe));
238 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
239 }
240 else
241 sfp = (struct sigframe *)regs->tf_esp - 1;
242
243 /* Translate the signal is appropriate */
244 if (p->p_sysent->sv_sigtbl) {
245 if (sig <= p->p_sysent->sv_sigsize)
246 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)];
247 }
248
249 /* Build the argument list for the signal handler. */
250 sf.sf_signum = sig;
251 sf.sf_ucontext = (register_t)&sfp->sf_uc;
252 if (SIGISMEMBER(psp->ps_siginfo, sig)) {
253 /* Signal handler installed with SA_SIGINFO. */
254 sf.sf_siginfo = (register_t)&sfp->sf_si;
255 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher;
256
257 /* fill siginfo structure */
258 sf.sf_si.si_signo = sig;
259 sf.sf_si.si_code = code;
260 sf.sf_si.si_addr = (void*)regs->tf_err;
261 }
262 else {
263 /* Old FreeBSD-style arguments. */
264 sf.sf_siginfo = code;
265 sf.sf_addr = regs->tf_err;
266 sf.sf_ahu.sf_handler = catcher;
267 }
268
269#if 0
270 /*
271 * If we're a vm86 process, we want to save the segment registers.
272 * We also change eflags to be our emulated eflags, not the actual
273 * eflags.
274 */
275 if (regs->tf_eflags & PSL_VM) {
276 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
277 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
278
279 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs;
280 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs;
281 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es;
282 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds;
283
284 if (vm86->vm86_has_vme == 0)
285 sf.sf_uc.uc_mcontext.mc_eflags =
286 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) |
287 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP));
288
289 /*
290 * Clear PSL_NT to inhibit T_TSSFLT faults on return from
291 * syscalls made by the signal handler. This just avoids
292 * wasting time for our lazy fixup of such faults. PSL_NT
293 * does nothing in vm86 mode, but vm86 programs can set it
294 * almost legitimately in probes for old cpu types.
295 */
296 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP);
297 }
298#endif
299
300 /*
301 * Copy the sigframe out to the user's stack.
302 */
303 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) {
304 /*
305 * Something is wrong with the stack pointer.
306 * ...Kill the process.
307 */
308 sigexit(p, SIGILL);
309 }
310
311 regs->tf_esp = (int)sfp;
312 regs->tf_eip = PS_STRINGS - *(p->p_sysent->sv_szsigcode);
313 regs->tf_eflags &= ~PSL_T;
4e7c41c5
MD
314 regs->tf_cs = _ucodesel;
315 regs->tf_ds = _udatasel;
316 regs->tf_es = _udatasel;
317 if (regs->tf_trapno == T_PROTFLT) {
318 regs->tf_fs = _udatasel;
319 regs->tf_gs = _udatasel;
320 }
321 regs->tf_ss = _udatasel;
6f7b98e0
MD
322}
323
324/*
325 * Sanitize the trapframe for a virtual kernel passing control to a custom
326 * VM context.
327 *
328 * Allow userland to set or maintain PSL_RF, the resume flag. This flag
329 * basically controls whether the return PC should skip the first instruction
330 * (as in an explicit system call) or re-execute it (as in an exception).
331 */
332int
333cpu_sanitize_frame(struct trapframe *frame)
334{
4e7c41c5
MD
335 frame->tf_cs = _ucodesel;
336 frame->tf_ds = _udatasel;
337 frame->tf_es = _udatasel;
338#if 0
339 frame->tf_fs = _udatasel;
340 frame->tf_gs = _udatasel;
341#endif
342 frame->tf_ss = _udatasel;
343 frame->tf_eflags &= (PSL_RF | PSL_USERCHANGE);
6f7b98e0
MD
344 frame->tf_eflags |= PSL_RESERVED_DEFAULT | PSL_I;
345 return(0);
346}
347
4e7c41c5
MD
348int
349cpu_sanitize_tls(struct savetls *tls)
350{
351 struct segment_descriptor *desc;
352 int i;
353
354 for (i = 0; i < NGTLS; ++i) {
355 desc = &tls->tls[i];
356 if (desc->sd_dpl == 0 && desc->sd_type == 0)
357 continue;
358 if (desc->sd_def32 == 0)
359 return(ENXIO);
360 if (desc->sd_type != SDT_MEMRWA)
361 return(ENXIO);
362 if (desc->sd_dpl != SEL_UPL)
363 return(ENXIO);
364 if (desc->sd_xx != 0 || desc->sd_p != 1)
365 return(ENXIO);
366 }
367 return(0);
368}
369
6f7b98e0
MD
370/*
371 * sigreturn(ucontext_t *sigcntxp)
372 *
373 * System call to cleanup state after a signal
374 * has been taken. Reset signal mask and
375 * stack state from context left by sendsig (above).
376 * Return to previous pc and psl as specified by
377 * context left by sendsig. Check carefully to
378 * make sure that the user has not modified the
379 * state to gain improper privileges.
380 */
381#define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
382#define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
383
384int
385sys_sigreturn(struct sigreturn_args *uap)
386{
387 struct lwp *lp = curthread->td_lwp;
388 struct trapframe *regs;
389 ucontext_t *ucp;
390 int cs, eflags;
391
392 ucp = uap->sigcntxp;
393
394 if (!useracc((caddr_t)ucp, sizeof(ucontext_t), VM_PROT_READ))
395 return (EFAULT);
396
397 regs = lp->lwp_md.md_regs;
398 eflags = ucp->uc_mcontext.mc_eflags;
399
400#if 0
401 if (eflags & PSL_VM) {
402 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs;
403 struct vm86_kernel *vm86;
404
405 /*
406 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't
407 * set up the vm86 area, and we can't enter vm86 mode.
408 */
409 if (lp->lwp_thread->td_pcb->pcb_ext == 0)
410 return (EINVAL);
411 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86;
412 if (vm86->vm86_inited == 0)
413 return (EINVAL);
414
415 /* go back to user mode if both flags are set */
416 if ((eflags & PSL_VIP) && (eflags & PSL_VIF))
417 trapsignal(lp->lwp_proc, SIGBUS, 0);
418
419 if (vm86->vm86_has_vme) {
420 eflags = (tf->tf_eflags & ~VME_USERCHANGE) |
421 (eflags & VME_USERCHANGE) | PSL_VM;
422 } else {
423 vm86->vm86_eflags = eflags; /* save VIF, VIP */
424 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM;
425 }
4e7c41c5 426 bcopy(&ucp->uc_mcontext.mc_gs, tf, sizeof(struct trapframe));
6f7b98e0
MD
427 tf->tf_eflags = eflags;
428 tf->tf_vm86_ds = tf->tf_ds;
429 tf->tf_vm86_es = tf->tf_es;
430 tf->tf_vm86_fs = tf->tf_fs;
4e7c41c5
MD
431 tf->tf_vm86_gs = tf->tf_gs;
432 tf->tf_ds = _udatasel;
433 tf->tf_es = _udatasel;
434#if 0
435 tf->tf_fs = _udatasel;
436 tf->tf_gs = _udatasel;
437#endif
6f7b98e0
MD
438 } else
439#endif
440 {
441 /*
442 * Don't allow users to change privileged or reserved flags.
443 */
444 /*
445 * XXX do allow users to change the privileged flag PSL_RF.
446 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers
447 * should sometimes set it there too. tf_eflags is kept in
448 * the signal context during signal handling and there is no
449 * other place to remember it, so the PSL_RF bit may be
450 * corrupted by the signal handler without us knowing.
451 * Corruption of the PSL_RF bit at worst causes one more or
452 * one less debugger trap, so allowing it is fairly harmless.
453 */
454 if (!EFL_SECURE(eflags & ~PSL_RF, regs->tf_eflags & ~PSL_RF)) {
455 kprintf("sigreturn: eflags = 0x%x\n", eflags);
456 return(EINVAL);
457 }
458
459 /*
460 * Don't allow users to load a valid privileged %cs. Let the
461 * hardware check for invalid selectors, excess privilege in
462 * other selectors, invalid %eip's and invalid %esp's.
463 */
464 cs = ucp->uc_mcontext.mc_cs;
465 if (!CS_SECURE(cs)) {
466 kprintf("sigreturn: cs = 0x%x\n", cs);
467 trapsignal(lp->lwp_proc, SIGBUS, T_PROTFLT);
468 return(EINVAL);
469 }
4e7c41c5 470 bcopy(&ucp->uc_mcontext.mc_gs, regs, sizeof(struct trapframe));
6f7b98e0
MD
471 }
472
473 if (ucp->uc_mcontext.mc_onstack & 1)
474 lp->lwp_sigstk.ss_flags |= SS_ONSTACK;
475 else
476 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK;
477
478 lp->lwp_sigmask = ucp->uc_sigmask;
479 SIG_CANTMASK(lp->lwp_sigmask);
480 return(EJUSTRETURN);
481}
482
483/*
484 * Stack frame on entry to function. %eax will contain the function vector,
485 * %ecx will contain the function data. flags, ecx, and eax will have
486 * already been pushed on the stack.
487 */
488struct upc_frame {
489 register_t eax;
490 register_t ecx;
491 register_t edx;
492 register_t flags;
493 register_t oldip;
494};
495
496void
497sendupcall(struct vmupcall *vu, int morepending)
498{
499 struct lwp *lp = curthread->td_lwp;
500 struct trapframe *regs;
501 struct upcall upcall;
502 struct upc_frame upc_frame;
503 int crit_count = 0;
504
505 /*
506 * Get the upcall data structure
507 */
508 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) ||
509 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int))
510 ) {
511 vu->vu_pending = 0;
512 kprintf("bad upcall address\n");
513 return;
514 }
515
516 /*
517 * If the data structure is already marked pending or has a critical
518 * section count, mark the data structure as pending and return
519 * without doing an upcall. vu_pending is left set.
520 */
521 if (upcall.upc_pending || crit_count >= vu->vu_pending) {
522 if (upcall.upc_pending < vu->vu_pending) {
523 upcall.upc_pending = vu->vu_pending;
524 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
525 sizeof(upcall.upc_pending));
526 }
527 return;
528 }
529
530 /*
531 * We can run this upcall now, clear vu_pending.
532 *
533 * Bump our critical section count and set or clear the
534 * user pending flag depending on whether more upcalls are
535 * pending. The user will be responsible for calling
536 * upc_dispatch(-1) to process remaining upcalls.
537 */
538 vu->vu_pending = 0;
539 upcall.upc_pending = morepending;
540 crit_count += TDPRI_CRIT;
541 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending,
542 sizeof(upcall.upc_pending));
543 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff,
544 sizeof(int));
545
546 /*
547 * Construct a stack frame and issue the upcall
548 */
549 regs = lp->lwp_md.md_regs;
550 upc_frame.eax = regs->tf_eax;
551 upc_frame.ecx = regs->tf_ecx;
552 upc_frame.edx = regs->tf_edx;
553 upc_frame.flags = regs->tf_eflags;
554 upc_frame.oldip = regs->tf_eip;
555 if (copyout(&upc_frame, (void *)(regs->tf_esp - sizeof(upc_frame)),
556 sizeof(upc_frame)) != 0) {
557 kprintf("bad stack on upcall\n");
558 } else {
559 regs->tf_eax = (register_t)vu->vu_func;
560 regs->tf_ecx = (register_t)vu->vu_data;
561 regs->tf_edx = (register_t)lp->lwp_upcall;
562 regs->tf_eip = (register_t)vu->vu_ctx;
563 regs->tf_esp -= sizeof(upc_frame);
564 }
565}
566
567/*
568 * fetchupcall occurs in the context of a system call, which means that
569 * we have to return EJUSTRETURN in order to prevent eax and edx from
570 * being overwritten by the syscall return value.
571 *
572 * if vu is not NULL we return the new context in %edx, the new data in %ecx,
573 * and the function pointer in %eax.
574 */
575int
576fetchupcall (struct vmupcall *vu, int morepending, void *rsp)
577{
578 struct upc_frame upc_frame;
579 struct lwp *lp = curthread->td_lwp;
580 struct trapframe *regs;
581 int error;
582 struct upcall upcall;
583 int crit_count;
584
585 regs = lp->lwp_md.md_regs;
586
587 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int));
588 if (error == 0) {
589 if (vu) {
590 /*
591 * This jumps us to the next ready context.
592 */
593 vu->vu_pending = 0;
594 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall));
595 crit_count = 0;
596 if (error == 0)
597 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int));
598 crit_count += TDPRI_CRIT;
599 if (error == 0)
600 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int));
601 regs->tf_eax = (register_t)vu->vu_func;
602 regs->tf_ecx = (register_t)vu->vu_data;
603 regs->tf_edx = (register_t)lp->lwp_upcall;
604 regs->tf_eip = (register_t)vu->vu_ctx;
605 regs->tf_esp = (register_t)rsp;
606 } else {
607 /*
608 * This returns us to the originally interrupted code.
609 */
610 error = copyin(rsp, &upc_frame, sizeof(upc_frame));
611 regs->tf_eax = upc_frame.eax;
612 regs->tf_ecx = upc_frame.ecx;
613 regs->tf_edx = upc_frame.edx;
614 regs->tf_eflags = (regs->tf_eflags & ~PSL_USERCHANGE) |
615 (upc_frame.flags & PSL_USERCHANGE);
616 regs->tf_eip = upc_frame.oldip;
617 regs->tf_esp = (register_t)((char *)rsp + sizeof(upc_frame));
618 }
619 }
620 if (error == 0)
621 error = EJUSTRETURN;
622 return(error);
623}
624
625/*
626 * cpu_idle() represents the idle LWKT. You cannot return from this function
627 * (unless you want to blow things up!). Instead we look for runnable threads
628 * and loop or halt as appropriate. Giant is not held on entry to the thread.
629 *
630 * The main loop is entered with a critical section held, we must release
631 * the critical section before doing anything else. lwkt_switch() will
632 * check for pending interrupts due to entering and exiting its own
633 * critical section.
634 *
635 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI
636 * to wake a HLTed cpu up. However, there are cases where the idlethread
637 * will be entered with the possibility that no IPI will occur and in such
638 * cases lwkt_switch() sets TDF_IDLE_NOHLT.
639 */
640static int cpu_idle_hlt = 1;
641static int cpu_idle_hltcnt;
642static int cpu_idle_spincnt;
643SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW,
644 &cpu_idle_hlt, 0, "Idle loop HLT enable");
645SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW,
646 &cpu_idle_hltcnt, 0, "Idle loop entry halts");
647SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW,
648 &cpu_idle_spincnt, 0, "Idle loop entry spins");
649
6f7b98e0
MD
650void
651cpu_idle(void)
652{
653 struct thread *td = curthread;
654
655 crit_exit();
656 KKASSERT(td->td_pri < TDPRI_CRIT);
657 for (;;) {
658 /*
659 * See if there are any LWKTs ready to go.
660 */
661 lwkt_switch();
662
663 /*
664 * If we are going to halt call splz unconditionally after
665 * CLIing to catch any interrupt races. Note that we are
666 * at SPL0 and interrupts are enabled.
667 */
668 if (cpu_idle_hlt && !lwkt_runnable() &&
669 (td->td_flags & TDF_IDLE_NOHLT) == 0) {
4e7c41c5 670 sigblock(SIGALRM);
6f7b98e0 671 splz();
4e7c41c5
MD
672 if (!lwkt_runnable()) {
673 sigpause(0);
674 } else {
675 sigblock(0);
676 }
6f7b98e0 677#ifdef SMP
4e7c41c5 678 else {
6f7b98e0 679 __asm __volatile("pause");
4e7c41c5 680 }
6f7b98e0
MD
681#endif
682 ++cpu_idle_hltcnt;
683 } else {
684 td->td_flags &= ~TDF_IDLE_NOHLT;
685 splz();
686#ifdef SMP
687 /*__asm __volatile("sti; pause");*/
688 __asm __volatile("pause");
689#else
690 /*__asm __volatile("sti");*/
691#endif
692 ++cpu_idle_spincnt;
693 }
694 }
695}
696
697/*
698 * Clear registers on exec
699 */
700void
701setregs(struct lwp *lp, u_long entry, u_long stack, u_long ps_strings)
702{
703 struct trapframe *regs = lp->lwp_md.md_regs;
704 struct pcb *pcb = lp->lwp_thread->td_pcb;
705
6f7b98e0
MD
706 /* was i386_user_cleanup() in NetBSD */
707 user_ldt_free(pcb);
708
709 bzero((char *)regs, sizeof(struct trapframe));
710 regs->tf_eip = entry;
711 regs->tf_esp = stack;
712 regs->tf_eflags = PSL_USER | (regs->tf_eflags & PSL_T);
713 regs->tf_ss = 0;
714 regs->tf_ds = 0;
715 regs->tf_es = 0;
716 regs->tf_fs = 0;
4e7c41c5 717 regs->tf_gs = 0;
6f7b98e0
MD
718 regs->tf_cs = 0;
719
720 /* PS_STRINGS value for BSD/OS binaries. It is 0 for non-BSD/OS. */
721 regs->tf_ebx = ps_strings;
722
723 /*
724 * Reset the hardware debug registers if they were in use.
725 * They won't have any meaning for the newly exec'd process.
726 */
727 if (pcb->pcb_flags & PCB_DBREGS) {
728 pcb->pcb_dr0 = 0;
729 pcb->pcb_dr1 = 0;
730 pcb->pcb_dr2 = 0;
731 pcb->pcb_dr3 = 0;
732 pcb->pcb_dr6 = 0;
733 pcb->pcb_dr7 = 0;
734 if (pcb == curthread->td_pcb) {
735 /*
736 * Clear the debug registers on the running
737 * CPU, otherwise they will end up affecting
738 * the next process we switch to.
739 */
740 reset_dbregs();
741 }
742 pcb->pcb_flags &= ~PCB_DBREGS;
743 }
744
745 /*
746 * Initialize the math emulator (if any) for the current process.
747 * Actually, just clear the bit that says that the emulator has
748 * been initialized. Initialization is delayed until the process
749 * traps to the emulator (if it is done at all) mainly because
750 * emulators don't provide an entry point for initialization.
751 */
752 lp->lwp_thread->td_pcb->pcb_flags &= ~FP_SOFTFP;
753
754 /*
755 * note: do not set CR0_TS here. npxinit() must do it after clearing
756 * gd_npxthread. Otherwise a preemptive interrupt thread may panic
757 * in npxdna().
758 */
759 crit_enter();
760#if 0
761 load_cr0(rcr0() | CR0_MP);
762#endif
763
764#if NNPX > 0
765 /* Initialize the npx (if any) for the current process. */
766 npxinit(__INITIAL_NPXCW__);
767#endif
768 crit_exit();
769
770 /*
771 * note: linux emulator needs edx to be 0x0 on entry, which is
772 * handled in execve simply by setting the 64 bit syscall
773 * return value to 0.
774 */
775}
776
777void
778cpu_setregs(void)
779{
780#if 0
781 unsigned int cr0;
782
783 cr0 = rcr0();
784 cr0 |= CR0_NE; /* Done by npxinit() */
785 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */
786#ifdef I386_CPU
787 if (cpu_class != CPUCLASS_386)
788#endif
789 cr0 |= CR0_WP | CR0_AM;
790 load_cr0(cr0);
791 load_gs(_udatasel);
792#endif
793}
794
795static int
796sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS)
797{
798 int error;
799 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2,
800 req);
801 if (!error && req->newptr)
802 resettodr();
803 return (error);
804}
805
806SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW,
807 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", "");
808
809extern u_long bootdev; /* not a cdev_t - encoding is different */
810SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev,
811 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)");
812
813/*
814 * Initialize 386 and configure to run kernel
815 */
816
817/*
818 * Initialize segments & interrupt table
819 */
820
821extern struct user *proc0paddr;
822
823#if 0
824
825extern inthand_t
826 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl),
827 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm),
828 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot),
829 IDTVEC(page), IDTVEC(mchk), IDTVEC(fpu), IDTVEC(align),
830 IDTVEC(xmm), IDTVEC(syscall),
831 IDTVEC(rsvd0);
832extern inthand_t
833 IDTVEC(int0x80_syscall);
834
835#endif
836
837#ifdef DEBUG_INTERRUPTS
838extern inthand_t *Xrsvdary[256];
839#endif
840
841int
842ptrace_set_pc(struct proc *p, unsigned long addr)
843{
844 p->p_md.md_regs->tf_eip = addr;
845 return (0);
846}
847
848int
849ptrace_single_step(struct lwp *lp)
850{
851 lp->lwp_md.md_regs->tf_eflags |= PSL_T;
852 return (0);
853}
854
855int
856fill_regs(struct lwp *lp, struct reg *regs)
857{
6f7b98e0
MD
858 struct trapframe *tp;
859
860 tp = lp->lwp_md.md_regs;
4e7c41c5 861 regs->r_gs = tp->tf_gs;
6f7b98e0
MD
862 regs->r_fs = tp->tf_fs;
863 regs->r_es = tp->tf_es;
864 regs->r_ds = tp->tf_ds;
865 regs->r_edi = tp->tf_edi;
866 regs->r_esi = tp->tf_esi;
867 regs->r_ebp = tp->tf_ebp;
868 regs->r_ebx = tp->tf_ebx;
869 regs->r_edx = tp->tf_edx;
870 regs->r_ecx = tp->tf_ecx;
871 regs->r_eax = tp->tf_eax;
872 regs->r_eip = tp->tf_eip;
873 regs->r_cs = tp->tf_cs;
874 regs->r_eflags = tp->tf_eflags;
875 regs->r_esp = tp->tf_esp;
876 regs->r_ss = tp->tf_ss;
6f7b98e0
MD
877 return (0);
878}
879
880int
881set_regs(struct lwp *lp, struct reg *regs)
882{
6f7b98e0
MD
883 struct trapframe *tp;
884
885 tp = lp->lwp_md.md_regs;
886 if (!EFL_SECURE(regs->r_eflags, tp->tf_eflags) ||
887 !CS_SECURE(regs->r_cs))
888 return (EINVAL);
4e7c41c5 889 tp->tf_gs = regs->r_gs;
6f7b98e0
MD
890 tp->tf_fs = regs->r_fs;
891 tp->tf_es = regs->r_es;
892 tp->tf_ds = regs->r_ds;
893 tp->tf_edi = regs->r_edi;
894 tp->tf_esi = regs->r_esi;
895 tp->tf_ebp = regs->r_ebp;
896 tp->tf_ebx = regs->r_ebx;
897 tp->tf_edx = regs->r_edx;
898 tp->tf_ecx = regs->r_ecx;
899 tp->tf_eax = regs->r_eax;
900 tp->tf_eip = regs->r_eip;
901 tp->tf_cs = regs->r_cs;
902 tp->tf_eflags = regs->r_eflags;
903 tp->tf_esp = regs->r_esp;
904 tp->tf_ss = regs->r_ss;
6f7b98e0
MD
905 return (0);
906}
907
908#ifndef CPU_DISABLE_SSE
909static void
910fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87)
911{
912 struct env87 *penv_87 = &sv_87->sv_env;
913 struct envxmm *penv_xmm = &sv_xmm->sv_env;
914 int i;
915
916 /* FPU control/status */
917 penv_87->en_cw = penv_xmm->en_cw;
918 penv_87->en_sw = penv_xmm->en_sw;
919 penv_87->en_tw = penv_xmm->en_tw;
920 penv_87->en_fip = penv_xmm->en_fip;
921 penv_87->en_fcs = penv_xmm->en_fcs;
922 penv_87->en_opcode = penv_xmm->en_opcode;
923 penv_87->en_foo = penv_xmm->en_foo;
924 penv_87->en_fos = penv_xmm->en_fos;
925
926 /* FPU registers */
927 for (i = 0; i < 8; ++i)
928 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc;
929
930 sv_87->sv_ex_sw = sv_xmm->sv_ex_sw;
931}
932
933static void
934set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm)
935{
936 struct env87 *penv_87 = &sv_87->sv_env;
937 struct envxmm *penv_xmm = &sv_xmm->sv_env;
938 int i;
939
940 /* FPU control/status */
941 penv_xmm->en_cw = penv_87->en_cw;
942 penv_xmm->en_sw = penv_87->en_sw;
943 penv_xmm->en_tw = penv_87->en_tw;
944 penv_xmm->en_fip = penv_87->en_fip;
945 penv_xmm->en_fcs = penv_87->en_fcs;
946 penv_xmm->en_opcode = penv_87->en_opcode;
947 penv_xmm->en_foo = penv_87->en_foo;
948 penv_xmm->en_fos = penv_87->en_fos;
949
950 /* FPU registers */
951 for (i = 0; i < 8; ++i)
952 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i];
953
954 sv_xmm->sv_ex_sw = sv_87->sv_ex_sw;
955}
956#endif /* CPU_DISABLE_SSE */
957
958int
959fill_fpregs(struct lwp *lp, struct fpreg *fpregs)
960{
961#ifndef CPU_DISABLE_SSE
962 if (cpu_fxsr) {
963 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm,
964 (struct save87 *)fpregs);
965 return (0);
966 }
967#endif /* CPU_DISABLE_SSE */
968 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs);
969 return (0);
970}
971
972int
973set_fpregs(struct lwp *lp, struct fpreg *fpregs)
974{
975#ifndef CPU_DISABLE_SSE
976 if (cpu_fxsr) {
977 set_fpregs_xmm((struct save87 *)fpregs,
978 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm);
979 return (0);
980 }
981#endif /* CPU_DISABLE_SSE */
982 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs);
983 return (0);
984}
985
986int
987fill_dbregs(struct lwp *lp, struct dbreg *dbregs)
988{
989 if (lp == NULL) {
990 dbregs->dr0 = rdr0();
991 dbregs->dr1 = rdr1();
992 dbregs->dr2 = rdr2();
993 dbregs->dr3 = rdr3();
994 dbregs->dr4 = rdr4();
995 dbregs->dr5 = rdr5();
996 dbregs->dr6 = rdr6();
997 dbregs->dr7 = rdr7();
998 } else {
999 struct pcb *pcb;
1000
1001 pcb = lp->lwp_thread->td_pcb;
1002 dbregs->dr0 = pcb->pcb_dr0;
1003 dbregs->dr1 = pcb->pcb_dr1;
1004 dbregs->dr2 = pcb->pcb_dr2;
1005 dbregs->dr3 = pcb->pcb_dr3;
1006 dbregs->dr4 = 0;
1007 dbregs->dr5 = 0;
1008 dbregs->dr6 = pcb->pcb_dr6;
1009 dbregs->dr7 = pcb->pcb_dr7;
1010 }
1011 return (0);
1012}
1013
1014int
1015set_dbregs(struct lwp *lp, struct dbreg *dbregs)
1016{
1017 if (lp == NULL) {
1018 load_dr0(dbregs->dr0);
1019 load_dr1(dbregs->dr1);
1020 load_dr2(dbregs->dr2);
1021 load_dr3(dbregs->dr3);
1022 load_dr4(dbregs->dr4);
1023 load_dr5(dbregs->dr5);
1024 load_dr6(dbregs->dr6);
1025 load_dr7(dbregs->dr7);
1026 } else {
1027 struct pcb *pcb;
1028 struct ucred *ucred;
1029 int i;
1030 uint32_t mask1, mask2;
1031
1032 /*
1033 * Don't let an illegal value for dr7 get set. Specifically,
1034 * check for undefined settings. Setting these bit patterns
1035 * result in undefined behaviour and can lead to an unexpected
1036 * TRCTRAP.
1037 */
1038 for (i = 0, mask1 = 0x3<<16, mask2 = 0x2<<16; i < 8;
1039 i++, mask1 <<= 2, mask2 <<= 2)
1040 if ((dbregs->dr7 & mask1) == mask2)
1041 return (EINVAL);
1042
1043 pcb = lp->lwp_thread->td_pcb;
1044 ucred = lp->lwp_proc->p_ucred;
1045
1046 /*
1047 * Don't let a process set a breakpoint that is not within the
1048 * process's address space. If a process could do this, it
1049 * could halt the system by setting a breakpoint in the kernel
1050 * (if ddb was enabled). Thus, we need to check to make sure
1051 * that no breakpoints are being enabled for addresses outside
1052 * process's address space, unless, perhaps, we were called by
1053 * uid 0.
1054 *
1055 * XXX - what about when the watched area of the user's
1056 * address space is written into from within the kernel
1057 * ... wouldn't that still cause a breakpoint to be generated
1058 * from within kernel mode?
1059 */
1060
1061 if (suser_cred(ucred, 0) != 0) {
1062 if (dbregs->dr7 & 0x3) {
1063 /* dr0 is enabled */
1064 if (dbregs->dr0 >= VM_MAX_USER_ADDRESS)
1065 return (EINVAL);
1066 }
1067
1068 if (dbregs->dr7 & (0x3<<2)) {
1069 /* dr1 is enabled */
1070 if (dbregs->dr1 >= VM_MAX_USER_ADDRESS)
1071 return (EINVAL);
1072 }
1073
1074 if (dbregs->dr7 & (0x3<<4)) {
1075 /* dr2 is enabled */
1076 if (dbregs->dr2 >= VM_MAX_USER_ADDRESS)
1077 return (EINVAL);
1078 }
1079
1080 if (dbregs->dr7 & (0x3<<6)) {
1081 /* dr3 is enabled */
1082 if (dbregs->dr3 >= VM_MAX_USER_ADDRESS)
1083 return (EINVAL);
1084 }
1085 }
1086
1087 pcb->pcb_dr0 = dbregs->dr0;
1088 pcb->pcb_dr1 = dbregs->dr1;
1089 pcb->pcb_dr2 = dbregs->dr2;
1090 pcb->pcb_dr3 = dbregs->dr3;
1091 pcb->pcb_dr6 = dbregs->dr6;
1092 pcb->pcb_dr7 = dbregs->dr7;
1093
1094 pcb->pcb_flags |= PCB_DBREGS;
1095 }
1096
1097 return (0);
1098}
1099
1100#if 0
1101/*
1102 * Return > 0 if a hardware breakpoint has been hit, and the
1103 * breakpoint was in user space. Return 0, otherwise.
1104 */
1105int
1106user_dbreg_trap(void)
1107{
1108 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */
1109 u_int32_t bp; /* breakpoint bits extracted from dr6 */
1110 int nbp; /* number of breakpoints that triggered */
1111 caddr_t addr[4]; /* breakpoint addresses */
1112 int i;
1113
1114 dr7 = rdr7();
1115 if ((dr7 & 0x000000ff) == 0) {
1116 /*
1117 * all GE and LE bits in the dr7 register are zero,
1118 * thus the trap couldn't have been caused by the
1119 * hardware debug registers
1120 */
1121 return 0;
1122 }
1123
1124 nbp = 0;
1125 dr6 = rdr6();
1126 bp = dr6 & 0x0000000f;
1127
1128 if (!bp) {
1129 /*
1130 * None of the breakpoint bits are set meaning this
1131 * trap was not caused by any of the debug registers
1132 */
1133 return 0;
1134 }
1135
1136 /*
1137 * at least one of the breakpoints were hit, check to see
1138 * which ones and if any of them are user space addresses
1139 */
1140
1141 if (bp & 0x01) {
1142 addr[nbp++] = (caddr_t)rdr0();
1143 }
1144 if (bp & 0x02) {
1145 addr[nbp++] = (caddr_t)rdr1();
1146 }
1147 if (bp & 0x04) {
1148 addr[nbp++] = (caddr_t)rdr2();
1149 }
1150 if (bp & 0x08) {
1151 addr[nbp++] = (caddr_t)rdr3();
1152 }
1153
1154 for (i=0; i<nbp; i++) {
1155 if (addr[i] <
1156 (caddr_t)VM_MAX_USER_ADDRESS) {
1157 /*
1158 * addr[i] is in user space
1159 */
1160 return nbp;
1161 }
1162 }
1163
1164 /*
1165 * None of the breakpoints are in user space.
1166 */
1167 return 0;
1168}
1169
1170#endif
1171
1172
1173#ifndef DDB
1174void
1175Debugger(const char *msg)
1176{
1177 kprintf("Debugger(\"%s\") called.\n", msg);
1178}
1179#endif /* no DDB */
1180
1181#include <sys/disklabel.h>
1182
1183/*
1184 * Determine the size of the transfer, and make sure it is
1185 * within the boundaries of the partition. Adjust transfer
1186 * if needed, and signal errors or early completion.
1187 *
1188 * On success a new bio layer is pushed with the translated
1189 * block number, and returned.
1190 */
1191struct bio *
1192bounds_check_with_label(cdev_t dev, struct bio *bio,
1193 struct disklabel *lp, int wlabel)
1194{
1195 struct bio *nbio;
1196 struct buf *bp = bio->bio_buf;
1197 struct partition *p = lp->d_partitions + dkpart(dev);
1198 int labelsect = lp->d_partitions[0].p_offset;
1199 int maxsz = p->p_size,
1200 sz = (bp->b_bcount + DEV_BSIZE - 1) >> DEV_BSHIFT;
1201 daddr_t blkno = (daddr_t)(bio->bio_offset >> DEV_BSHIFT);
1202
1203 /* overwriting disk label ? */
1204 /* XXX should also protect bootstrap in first 8K */
1205 if (blkno + p->p_offset <= LABELSECTOR + labelsect &&
1206#if LABELSECTOR != 0
1207 blkno + p->p_offset + sz > LABELSECTOR + labelsect &&
1208#endif
1209 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1210 bp->b_error = EROFS;
1211 goto error;
1212 }
1213
1214#if defined(DOSBBSECTOR) && defined(notyet)
1215 /* overwriting master boot record? */
1216 if (blkno + p->p_offset <= DOSBBSECTOR &&
1217 bp->b_cmd != BUF_CMD_READ && wlabel == 0) {
1218 bp->b_error = EROFS;
1219 goto error;
1220 }
1221#endif
1222
1223 /*
1224 * Check for out of bounds, EOF, and EOF clipping.
1225 */
1226 if (bio->bio_offset < 0)
1227 goto bad;
1228 if (blkno + sz > maxsz) {
1229 /*
1230 * Past EOF or B_BNOCLIP flag was set, the request is bad.
1231 */
1232 if (blkno > maxsz || (bp->b_flags & B_BNOCLIP))
1233 goto bad;
1234
1235 /*
1236 * If exactly on EOF just complete the I/O with no bytes
1237 * transfered. B_INVAL must be set to throw away the
1238 * contents of the buffer. Otherwise clip b_bcount.
1239 */
1240 if (blkno == maxsz) {
1241 bp->b_resid = bp->b_bcount;
1242 bp->b_flags |= B_INVAL;
1243 goto done;
1244 }
1245 bp->b_bcount = (maxsz - blkno) << DEV_BSHIFT;
1246 }
1247 nbio = push_bio(bio);
1248 nbio->bio_offset = bio->bio_offset + ((off_t)p->p_offset << DEV_BSHIFT);
1249 return (nbio);
1250
1251 /*
1252 * The caller is responsible for calling biodone() on the passed bio
1253 * when we return NULL.
1254 */
1255bad:
1256 bp->b_error = EINVAL;
1257error:
1258 bp->b_resid = bp->b_bcount;
1259 bp->b_flags |= B_ERROR | B_INVAL;
1260done:
1261 return (NULL);
1262}
1263