| 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 $ |
| 40 | */ |
| 41 | |
| 42 | #include "opt_compat.h" |
| 43 | #include "opt_ddb.h" |
| 44 | #include "opt_directio.h" |
| 45 | #include "opt_inet.h" |
| 46 | #include "opt_msgbuf.h" |
| 47 | #include "opt_swap.h" |
| 48 | |
| 49 | #include <sys/param.h> |
| 50 | #include <sys/systm.h> |
| 51 | #include <sys/sysproto.h> |
| 52 | #include <sys/signalvar.h> |
| 53 | #include <sys/kernel.h> |
| 54 | #include <sys/linker.h> |
| 55 | #include <sys/malloc.h> |
| 56 | #include <sys/proc.h> |
| 57 | #include <sys/buf.h> |
| 58 | #include <sys/reboot.h> |
| 59 | #include <sys/mbuf.h> |
| 60 | #include <sys/msgbuf.h> |
| 61 | #include <sys/sysent.h> |
| 62 | #include <sys/sysctl.h> |
| 63 | #include <sys/vmmeter.h> |
| 64 | #include <sys/bus.h> |
| 65 | #include <sys/usched.h> |
| 66 | #include <sys/reg.h> |
| 67 | |
| 68 | #include <vm/vm.h> |
| 69 | #include <vm/vm_param.h> |
| 70 | #include <sys/lock.h> |
| 71 | #include <vm/vm_kern.h> |
| 72 | #include <vm/vm_object.h> |
| 73 | #include <vm/vm_page.h> |
| 74 | #include <vm/vm_map.h> |
| 75 | #include <vm/vm_pager.h> |
| 76 | #include <vm/vm_extern.h> |
| 77 | |
| 78 | #include <sys/thread2.h> |
| 79 | #include <sys/mplock2.h> |
| 80 | |
| 81 | #include <sys/user.h> |
| 82 | #include <sys/exec.h> |
| 83 | #include <sys/cons.h> |
| 84 | |
| 85 | #include <ddb/ddb.h> |
| 86 | |
| 87 | #include <machine/cpu.h> |
| 88 | #include <machine/clock.h> |
| 89 | #include <machine/specialreg.h> |
| 90 | #include <machine/md_var.h> |
| 91 | #include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */ |
| 92 | #include <machine/globaldata.h> /* CPU_prvspace */ |
| 93 | #include <machine/smp.h> |
| 94 | #ifdef PERFMON |
| 95 | #include <machine/perfmon.h> |
| 96 | #endif |
| 97 | #include <machine/cputypes.h> |
| 98 | |
| 99 | #include <bus/isa/rtc.h> |
| 100 | #include <sys/random.h> |
| 101 | #include <sys/ptrace.h> |
| 102 | #include <machine/sigframe.h> |
| 103 | #include <unistd.h> /* umtx_* functions */ |
| 104 | #include <pthread.h> /* pthread_yield() */ |
| 105 | |
| 106 | extern void dblfault_handler (void); |
| 107 | |
| 108 | static void set_fpregs_xmm (struct save87 *, struct savexmm *); |
| 109 | static void fill_fpregs_xmm (struct savexmm *, struct save87 *); |
| 110 | #ifdef DIRECTIO |
| 111 | extern void ffs_rawread_setup(void); |
| 112 | #endif /* DIRECTIO */ |
| 113 | |
| 114 | int64_t tsc_offsets[MAXCPU]; |
| 115 | |
| 116 | #if defined(SWTCH_OPTIM_STATS) |
| 117 | extern int swtch_optim_stats; |
| 118 | SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats, |
| 119 | CTLFLAG_RD, &swtch_optim_stats, 0, ""); |
| 120 | SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count, |
| 121 | CTLFLAG_RD, &tlb_flush_count, 0, ""); |
| 122 | #endif |
| 123 | |
| 124 | static int |
| 125 | sysctl_hw_physmem(SYSCTL_HANDLER_ARGS) |
| 126 | { |
| 127 | u_long pmem = ctob(physmem); |
| 128 | |
| 129 | int error = sysctl_handle_long(oidp, &pmem, 0, req); |
| 130 | return (error); |
| 131 | } |
| 132 | |
| 133 | SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD, |
| 134 | 0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)"); |
| 135 | |
| 136 | static int |
| 137 | sysctl_hw_usermem(SYSCTL_HANDLER_ARGS) |
| 138 | { |
| 139 | /* JG */ |
| 140 | int error = sysctl_handle_int(oidp, 0, |
| 141 | ctob((int)Maxmem - vmstats.v_wire_count), req); |
| 142 | return (error); |
| 143 | } |
| 144 | |
| 145 | SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD, |
| 146 | 0, 0, sysctl_hw_usermem, "IU", ""); |
| 147 | |
| 148 | SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, ""); |
| 149 | |
| 150 | /* |
| 151 | * Send an interrupt to process. |
| 152 | * |
| 153 | * Stack is set up to allow sigcode stored |
| 154 | * at top to call routine, followed by kcall |
| 155 | * to sigreturn routine below. After sigreturn |
| 156 | * resets the signal mask, the stack, and the |
| 157 | * frame pointer, it returns to the user |
| 158 | * specified pc, psl. |
| 159 | */ |
| 160 | void |
| 161 | sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code) |
| 162 | { |
| 163 | struct lwp *lp = curthread->td_lwp; |
| 164 | struct proc *p = lp->lwp_proc; |
| 165 | struct trapframe *regs; |
| 166 | struct sigacts *psp = p->p_sigacts; |
| 167 | struct sigframe sf, *sfp; |
| 168 | int oonstack; |
| 169 | char *sp; |
| 170 | |
| 171 | regs = lp->lwp_md.md_regs; |
| 172 | oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0; |
| 173 | |
| 174 | /* Save user context */ |
| 175 | bzero(&sf, sizeof(struct sigframe)); |
| 176 | sf.sf_uc.uc_sigmask = *mask; |
| 177 | sf.sf_uc.uc_stack = lp->lwp_sigstk; |
| 178 | sf.sf_uc.uc_mcontext.mc_onstack = oonstack; |
| 179 | KKASSERT(__offsetof(struct trapframe, tf_rdi) == 0); |
| 180 | bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(struct trapframe)); |
| 181 | |
| 182 | /* Make the size of the saved context visible to userland */ |
| 183 | sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); |
| 184 | |
| 185 | /* Allocate and validate space for the signal handler context. */ |
| 186 | if ((lp->lwp_flags & LWP_ALTSTACK) != 0 && !oonstack && |
| 187 | SIGISMEMBER(psp->ps_sigonstack, sig)) { |
| 188 | sp = (char *)(lp->lwp_sigstk.ss_sp + lp->lwp_sigstk.ss_size - |
| 189 | sizeof(struct sigframe)); |
| 190 | lp->lwp_sigstk.ss_flags |= SS_ONSTACK; |
| 191 | } else { |
| 192 | /* We take red zone into account */ |
| 193 | sp = (char *)regs->tf_rsp - sizeof(struct sigframe) - 128; |
| 194 | } |
| 195 | |
| 196 | /* Align to 16 bytes */ |
| 197 | sfp = (struct sigframe *)((intptr_t)sp & ~0xFUL); |
| 198 | |
| 199 | /* Translate the signal is appropriate */ |
| 200 | if (p->p_sysent->sv_sigtbl) { |
| 201 | if (sig <= p->p_sysent->sv_sigsize) |
| 202 | sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)]; |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * Build the argument list for the signal handler. |
| 207 | * |
| 208 | * Arguments are in registers (%rdi, %rsi, %rdx, %rcx) |
| 209 | */ |
| 210 | regs->tf_rdi = sig; /* argument 1 */ |
| 211 | regs->tf_rdx = (register_t)&sfp->sf_uc; /* argument 3 */ |
| 212 | |
| 213 | if (SIGISMEMBER(psp->ps_siginfo, sig)) { |
| 214 | /* |
| 215 | * Signal handler installed with SA_SIGINFO. |
| 216 | * |
| 217 | * action(signo, siginfo, ucontext) |
| 218 | */ |
| 219 | regs->tf_rsi = (register_t)&sfp->sf_si; /* argument 2 */ |
| 220 | regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */ |
| 221 | sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher; |
| 222 | |
| 223 | /* fill siginfo structure */ |
| 224 | sf.sf_si.si_signo = sig; |
| 225 | sf.sf_si.si_code = code; |
| 226 | sf.sf_si.si_addr = (void *)regs->tf_addr; |
| 227 | } else { |
| 228 | /* |
| 229 | * Old FreeBSD-style arguments. |
| 230 | * |
| 231 | * handler (signo, code, [uc], addr) |
| 232 | */ |
| 233 | regs->tf_rsi = (register_t)code; /* argument 2 */ |
| 234 | regs->tf_rcx = (register_t)regs->tf_addr; /* argument 4 */ |
| 235 | sf.sf_ahu.sf_handler = catcher; |
| 236 | } |
| 237 | |
| 238 | #if 0 |
| 239 | /* |
| 240 | * If we're a vm86 process, we want to save the segment registers. |
| 241 | * We also change eflags to be our emulated eflags, not the actual |
| 242 | * eflags. |
| 243 | */ |
| 244 | if (regs->tf_eflags & PSL_VM) { |
| 245 | struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs; |
| 246 | struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86; |
| 247 | |
| 248 | sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs; |
| 249 | sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs; |
| 250 | sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es; |
| 251 | sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds; |
| 252 | |
| 253 | if (vm86->vm86_has_vme == 0) |
| 254 | sf.sf_uc.uc_mcontext.mc_eflags = |
| 255 | (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) | |
| 256 | (vm86->vm86_eflags & (PSL_VIF | PSL_VIP)); |
| 257 | |
| 258 | /* |
| 259 | * Clear PSL_NT to inhibit T_TSSFLT faults on return from |
| 260 | * syscalls made by the signal handler. This just avoids |
| 261 | * wasting time for our lazy fixup of such faults. PSL_NT |
| 262 | * does nothing in vm86 mode, but vm86 programs can set it |
| 263 | * almost legitimately in probes for old cpu types. |
| 264 | */ |
| 265 | tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP); |
| 266 | } |
| 267 | #endif |
| 268 | |
| 269 | /* |
| 270 | * Save the FPU state and reinit the FP unit |
| 271 | */ |
| 272 | npxpush(&sf.sf_uc.uc_mcontext); |
| 273 | |
| 274 | /* |
| 275 | * Copy the sigframe out to the user's stack. |
| 276 | */ |
| 277 | if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) { |
| 278 | /* |
| 279 | * Something is wrong with the stack pointer. |
| 280 | * ...Kill the process. |
| 281 | */ |
| 282 | sigexit(lp, SIGILL); |
| 283 | } |
| 284 | |
| 285 | regs->tf_rsp = (register_t)sfp; |
| 286 | regs->tf_rip = PS_STRINGS - *(p->p_sysent->sv_szsigcode); |
| 287 | |
| 288 | /* |
| 289 | * i386 abi specifies that the direction flag must be cleared |
| 290 | * on function entry |
| 291 | */ |
| 292 | regs->tf_rflags &= ~(PSL_T|PSL_D); |
| 293 | |
| 294 | /* |
| 295 | * 64 bit mode has a code and stack selector but |
| 296 | * no data or extra selector. %fs and %gs are not |
| 297 | * stored in-context. |
| 298 | */ |
| 299 | regs->tf_cs = _ucodesel; |
| 300 | regs->tf_ss = _udatasel; |
| 301 | } |
| 302 | |
| 303 | /* |
| 304 | * Sanitize the trapframe for a virtual kernel passing control to a custom |
| 305 | * VM context. Remove any items that would otherwise create a privilage |
| 306 | * issue. |
| 307 | * |
| 308 | * XXX at the moment we allow userland to set the resume flag. Is this a |
| 309 | * bad idea? |
| 310 | */ |
| 311 | int |
| 312 | cpu_sanitize_frame(struct trapframe *frame) |
| 313 | { |
| 314 | frame->tf_cs = _ucodesel; |
| 315 | frame->tf_ss = _udatasel; |
| 316 | /* XXX VM (8086) mode not supported? */ |
| 317 | frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE | PSL_VM_UNSUPP); |
| 318 | frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I; |
| 319 | |
| 320 | return(0); |
| 321 | } |
| 322 | |
| 323 | /* |
| 324 | * Sanitize the tls so loading the descriptor does not blow up |
| 325 | * on us. For x86_64 we don't have to do anything. |
| 326 | */ |
| 327 | int |
| 328 | cpu_sanitize_tls(struct savetls *tls) |
| 329 | { |
| 330 | return(0); |
| 331 | } |
| 332 | |
| 333 | /* |
| 334 | * sigreturn(ucontext_t *sigcntxp) |
| 335 | * |
| 336 | * System call to cleanup state after a signal |
| 337 | * has been taken. Reset signal mask and |
| 338 | * stack state from context left by sendsig (above). |
| 339 | * Return to previous pc and psl as specified by |
| 340 | * context left by sendsig. Check carefully to |
| 341 | * make sure that the user has not modified the |
| 342 | * state to gain improper privileges. |
| 343 | */ |
| 344 | #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) |
| 345 | #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) |
| 346 | |
| 347 | int |
| 348 | sys_sigreturn(struct sigreturn_args *uap) |
| 349 | { |
| 350 | struct lwp *lp = curthread->td_lwp; |
| 351 | struct trapframe *regs; |
| 352 | ucontext_t uc; |
| 353 | ucontext_t *ucp; |
| 354 | register_t rflags; |
| 355 | int cs; |
| 356 | int error; |
| 357 | |
| 358 | /* |
| 359 | * We have to copy the information into kernel space so userland |
| 360 | * can't modify it while we are sniffing it. |
| 361 | */ |
| 362 | regs = lp->lwp_md.md_regs; |
| 363 | error = copyin(uap->sigcntxp, &uc, sizeof(uc)); |
| 364 | if (error) |
| 365 | return (error); |
| 366 | ucp = &uc; |
| 367 | rflags = ucp->uc_mcontext.mc_rflags; |
| 368 | |
| 369 | /* VM (8086) mode not supported */ |
| 370 | rflags &= ~PSL_VM_UNSUPP; |
| 371 | |
| 372 | #if 0 |
| 373 | if (eflags & PSL_VM) { |
| 374 | struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs; |
| 375 | struct vm86_kernel *vm86; |
| 376 | |
| 377 | /* |
| 378 | * if pcb_ext == 0 or vm86_inited == 0, the user hasn't |
| 379 | * set up the vm86 area, and we can't enter vm86 mode. |
| 380 | */ |
| 381 | if (lp->lwp_thread->td_pcb->pcb_ext == 0) |
| 382 | return (EINVAL); |
| 383 | vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86; |
| 384 | if (vm86->vm86_inited == 0) |
| 385 | return (EINVAL); |
| 386 | |
| 387 | /* go back to user mode if both flags are set */ |
| 388 | if ((eflags & PSL_VIP) && (eflags & PSL_VIF)) |
| 389 | trapsignal(lp->lwp_proc, SIGBUS, 0); |
| 390 | |
| 391 | if (vm86->vm86_has_vme) { |
| 392 | eflags = (tf->tf_eflags & ~VME_USERCHANGE) | |
| 393 | (eflags & VME_USERCHANGE) | PSL_VM; |
| 394 | } else { |
| 395 | vm86->vm86_eflags = eflags; /* save VIF, VIP */ |
| 396 | eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM; |
| 397 | } |
| 398 | bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe)); |
| 399 | tf->tf_eflags = eflags; |
| 400 | tf->tf_vm86_ds = tf->tf_ds; |
| 401 | tf->tf_vm86_es = tf->tf_es; |
| 402 | tf->tf_vm86_fs = tf->tf_fs; |
| 403 | tf->tf_vm86_gs = tf->tf_gs; |
| 404 | tf->tf_ds = _udatasel; |
| 405 | tf->tf_es = _udatasel; |
| 406 | #if 0 |
| 407 | tf->tf_fs = _udatasel; |
| 408 | tf->tf_gs = _udatasel; |
| 409 | #endif |
| 410 | } else |
| 411 | #endif |
| 412 | { |
| 413 | /* |
| 414 | * Don't allow users to change privileged or reserved flags. |
| 415 | */ |
| 416 | /* |
| 417 | * XXX do allow users to change the privileged flag PSL_RF. |
| 418 | * The cpu sets PSL_RF in tf_eflags for faults. Debuggers |
| 419 | * should sometimes set it there too. tf_eflags is kept in |
| 420 | * the signal context during signal handling and there is no |
| 421 | * other place to remember it, so the PSL_RF bit may be |
| 422 | * corrupted by the signal handler without us knowing. |
| 423 | * Corruption of the PSL_RF bit at worst causes one more or |
| 424 | * one less debugger trap, so allowing it is fairly harmless. |
| 425 | */ |
| 426 | if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { |
| 427 | kprintf("sigreturn: rflags = 0x%lx\n", (long)rflags); |
| 428 | return(EINVAL); |
| 429 | } |
| 430 | |
| 431 | /* |
| 432 | * Don't allow users to load a valid privileged %cs. Let the |
| 433 | * hardware check for invalid selectors, excess privilege in |
| 434 | * other selectors, invalid %eip's and invalid %esp's. |
| 435 | */ |
| 436 | cs = ucp->uc_mcontext.mc_cs; |
| 437 | if (!CS_SECURE(cs)) { |
| 438 | kprintf("sigreturn: cs = 0x%x\n", cs); |
| 439 | trapsignal(lp, SIGBUS, T_PROTFLT); |
| 440 | return(EINVAL); |
| 441 | } |
| 442 | bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(struct trapframe)); |
| 443 | } |
| 444 | |
| 445 | /* |
| 446 | * Restore the FPU state from the frame |
| 447 | */ |
| 448 | npxpop(&ucp->uc_mcontext); |
| 449 | |
| 450 | if (ucp->uc_mcontext.mc_onstack & 1) |
| 451 | lp->lwp_sigstk.ss_flags |= SS_ONSTACK; |
| 452 | else |
| 453 | lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK; |
| 454 | |
| 455 | lp->lwp_sigmask = ucp->uc_sigmask; |
| 456 | SIG_CANTMASK(lp->lwp_sigmask); |
| 457 | return(EJUSTRETURN); |
| 458 | } |
| 459 | |
| 460 | /* |
| 461 | * cpu_idle() represents the idle LWKT. You cannot return from this function |
| 462 | * (unless you want to blow things up!). Instead we look for runnable threads |
| 463 | * and loop or halt as appropriate. Giant is not held on entry to the thread. |
| 464 | * |
| 465 | * The main loop is entered with a critical section held, we must release |
| 466 | * the critical section before doing anything else. lwkt_switch() will |
| 467 | * check for pending interrupts due to entering and exiting its own |
| 468 | * critical section. |
| 469 | * |
| 470 | * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI |
| 471 | * to wake a HLTed cpu up. |
| 472 | */ |
| 473 | static int cpu_idle_hlt = 1; |
| 474 | static int cpu_idle_hltcnt; |
| 475 | static int cpu_idle_spincnt; |
| 476 | SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW, |
| 477 | &cpu_idle_hlt, 0, "Idle loop HLT enable"); |
| 478 | SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW, |
| 479 | &cpu_idle_hltcnt, 0, "Idle loop entry halts"); |
| 480 | SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW, |
| 481 | &cpu_idle_spincnt, 0, "Idle loop entry spins"); |
| 482 | |
| 483 | void |
| 484 | cpu_idle(void) |
| 485 | { |
| 486 | struct thread *td = curthread; |
| 487 | struct mdglobaldata *gd = mdcpu; |
| 488 | int reqflags; |
| 489 | |
| 490 | crit_exit(); |
| 491 | KKASSERT(td->td_critcount == 0); |
| 492 | cpu_enable_intr(); |
| 493 | |
| 494 | for (;;) { |
| 495 | /* |
| 496 | * See if there are any LWKTs ready to go. |
| 497 | */ |
| 498 | lwkt_switch(); |
| 499 | |
| 500 | /* |
| 501 | * The idle loop halts only if no threads are scheduleable |
| 502 | * and no signals have occured. |
| 503 | */ |
| 504 | if (cpu_idle_hlt && |
| 505 | (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) { |
| 506 | splz(); |
| 507 | if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) { |
| 508 | #ifdef DEBUGIDLE |
| 509 | struct timeval tv1, tv2; |
| 510 | gettimeofday(&tv1, NULL); |
| 511 | #endif |
| 512 | reqflags = gd->mi.gd_reqflags & |
| 513 | ~RQF_IDLECHECK_WK_MASK; |
| 514 | KKASSERT(gd->mi.gd_processing_ipiq == 0); |
| 515 | umtx_sleep(&gd->mi.gd_reqflags, reqflags, |
| 516 | 1000000); |
| 517 | #ifdef DEBUGIDLE |
| 518 | gettimeofday(&tv2, NULL); |
| 519 | if (tv2.tv_usec - tv1.tv_usec + |
| 520 | (tv2.tv_sec - tv1.tv_sec) * 1000000 |
| 521 | > 500000) { |
| 522 | kprintf("cpu %d idlelock %08x %08x\n", |
| 523 | gd->mi.gd_cpuid, |
| 524 | gd->mi.gd_reqflags, |
| 525 | gd->gd_fpending); |
| 526 | } |
| 527 | #endif |
| 528 | } |
| 529 | ++cpu_idle_hltcnt; |
| 530 | } else { |
| 531 | splz(); |
| 532 | __asm __volatile("pause"); |
| 533 | ++cpu_idle_spincnt; |
| 534 | } |
| 535 | } |
| 536 | } |
| 537 | |
| 538 | /* |
| 539 | * Called by the spinlock code with or without a critical section held |
| 540 | * when a spinlock is found to be seriously constested. |
| 541 | * |
| 542 | * We need to enter a critical section to prevent signals from recursing |
| 543 | * into pthreads. |
| 544 | */ |
| 545 | void |
| 546 | cpu_spinlock_contested(void) |
| 547 | { |
| 548 | cpu_pause(); |
| 549 | } |
| 550 | |
| 551 | /* |
| 552 | * Clear registers on exec |
| 553 | */ |
| 554 | void |
| 555 | exec_setregs(u_long entry, u_long stack, u_long ps_strings) |
| 556 | { |
| 557 | struct thread *td = curthread; |
| 558 | struct lwp *lp = td->td_lwp; |
| 559 | struct pcb *pcb = td->td_pcb; |
| 560 | struct trapframe *regs = lp->lwp_md.md_regs; |
| 561 | |
| 562 | /* was i386_user_cleanup() in NetBSD */ |
| 563 | user_ldt_free(pcb); |
| 564 | |
| 565 | bzero((char *)regs, sizeof(struct trapframe)); |
| 566 | regs->tf_rip = entry; |
| 567 | regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; /* align the stack */ |
| 568 | regs->tf_rdi = stack; /* argv */ |
| 569 | regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); |
| 570 | regs->tf_ss = _udatasel; |
| 571 | regs->tf_cs = _ucodesel; |
| 572 | regs->tf_rbx = ps_strings; |
| 573 | |
| 574 | /* |
| 575 | * Reset the hardware debug registers if they were in use. |
| 576 | * They won't have any meaning for the newly exec'd process. |
| 577 | */ |
| 578 | if (pcb->pcb_flags & PCB_DBREGS) { |
| 579 | pcb->pcb_dr0 = 0; |
| 580 | pcb->pcb_dr1 = 0; |
| 581 | pcb->pcb_dr2 = 0; |
| 582 | pcb->pcb_dr3 = 0; |
| 583 | pcb->pcb_dr6 = 0; |
| 584 | pcb->pcb_dr7 = 0; /* JG set bit 10? */ |
| 585 | if (pcb == td->td_pcb) { |
| 586 | /* |
| 587 | * Clear the debug registers on the running |
| 588 | * CPU, otherwise they will end up affecting |
| 589 | * the next process we switch to. |
| 590 | */ |
| 591 | reset_dbregs(); |
| 592 | } |
| 593 | pcb->pcb_flags &= ~PCB_DBREGS; |
| 594 | } |
| 595 | |
| 596 | /* |
| 597 | * Initialize the math emulator (if any) for the current process. |
| 598 | * Actually, just clear the bit that says that the emulator has |
| 599 | * been initialized. Initialization is delayed until the process |
| 600 | * traps to the emulator (if it is done at all) mainly because |
| 601 | * emulators don't provide an entry point for initialization. |
| 602 | */ |
| 603 | pcb->pcb_flags &= ~FP_SOFTFP; |
| 604 | |
| 605 | /* |
| 606 | * NOTE: do not set CR0_TS here. npxinit() must do it after clearing |
| 607 | * gd_npxthread. Otherwise a preemptive interrupt thread |
| 608 | * may panic in npxdna(). |
| 609 | */ |
| 610 | crit_enter(); |
| 611 | #if 0 |
| 612 | load_cr0(rcr0() | CR0_MP); |
| 613 | #endif |
| 614 | |
| 615 | /* |
| 616 | * NOTE: The MSR values must be correct so we can return to |
| 617 | * userland. gd_user_fs/gs must be correct so the switch |
| 618 | * code knows what the current MSR values are. |
| 619 | */ |
| 620 | pcb->pcb_fsbase = 0; /* Values loaded from PCB on switch */ |
| 621 | pcb->pcb_gsbase = 0; |
| 622 | /* Initialize the npx (if any) for the current process. */ |
| 623 | npxinit(); |
| 624 | crit_exit(); |
| 625 | |
| 626 | /* |
| 627 | * note: linux emulator needs edx to be 0x0 on entry, which is |
| 628 | * handled in execve simply by setting the 64 bit syscall |
| 629 | * return value to 0. |
| 630 | */ |
| 631 | } |
| 632 | |
| 633 | void |
| 634 | cpu_setregs(void) |
| 635 | { |
| 636 | #if 0 |
| 637 | unsigned int cr0; |
| 638 | |
| 639 | cr0 = rcr0(); |
| 640 | cr0 |= CR0_NE; /* Done by npxinit() */ |
| 641 | cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */ |
| 642 | cr0 |= CR0_WP | CR0_AM; |
| 643 | load_cr0(cr0); |
| 644 | load_gs(_udatasel); |
| 645 | #endif |
| 646 | } |
| 647 | |
| 648 | static int |
| 649 | sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS) |
| 650 | { |
| 651 | int error; |
| 652 | error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, |
| 653 | req); |
| 654 | if (!error && req->newptr) |
| 655 | resettodr(); |
| 656 | return (error); |
| 657 | } |
| 658 | |
| 659 | SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW, |
| 660 | &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", ""); |
| 661 | |
| 662 | extern u_long bootdev; /* not a cdev_t - encoding is different */ |
| 663 | SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev, |
| 664 | CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)"); |
| 665 | |
| 666 | /* |
| 667 | * Initialize 386 and configure to run kernel |
| 668 | */ |
| 669 | |
| 670 | /* |
| 671 | * Initialize segments & interrupt table |
| 672 | */ |
| 673 | |
| 674 | extern struct user *proc0paddr; |
| 675 | |
| 676 | #if 0 |
| 677 | |
| 678 | extern inthand_t |
| 679 | IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl), |
| 680 | IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm), |
| 681 | IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot), |
| 682 | IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align), |
| 683 | IDTVEC(xmm), IDTVEC(dblfault), |
| 684 | IDTVEC(fast_syscall), IDTVEC(fast_syscall32); |
| 685 | #endif |
| 686 | |
| 687 | #ifdef DEBUG_INTERRUPTS |
| 688 | extern inthand_t *Xrsvdary[256]; |
| 689 | #endif |
| 690 | |
| 691 | int |
| 692 | ptrace_set_pc(struct lwp *lp, unsigned long addr) |
| 693 | { |
| 694 | lp->lwp_md.md_regs->tf_rip = addr; |
| 695 | return (0); |
| 696 | } |
| 697 | |
| 698 | int |
| 699 | ptrace_single_step(struct lwp *lp) |
| 700 | { |
| 701 | lp->lwp_md.md_regs->tf_rflags |= PSL_T; |
| 702 | return (0); |
| 703 | } |
| 704 | |
| 705 | int |
| 706 | fill_regs(struct lwp *lp, struct reg *regs) |
| 707 | { |
| 708 | struct trapframe *tp; |
| 709 | |
| 710 | if ((tp = lp->lwp_md.md_regs) == NULL) |
| 711 | return EINVAL; |
| 712 | bcopy(&tp->tf_rdi, ®s->r_rdi, sizeof(*regs)); |
| 713 | return (0); |
| 714 | } |
| 715 | |
| 716 | int |
| 717 | set_regs(struct lwp *lp, struct reg *regs) |
| 718 | { |
| 719 | struct trapframe *tp; |
| 720 | |
| 721 | tp = lp->lwp_md.md_regs; |
| 722 | if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) || |
| 723 | !CS_SECURE(regs->r_cs)) |
| 724 | return (EINVAL); |
| 725 | bcopy(®s->r_rdi, &tp->tf_rdi, sizeof(*regs)); |
| 726 | return (0); |
| 727 | } |
| 728 | |
| 729 | static void |
| 730 | fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87) |
| 731 | { |
| 732 | struct env87 *penv_87 = &sv_87->sv_env; |
| 733 | struct envxmm *penv_xmm = &sv_xmm->sv_env; |
| 734 | int i; |
| 735 | |
| 736 | /* FPU control/status */ |
| 737 | penv_87->en_cw = penv_xmm->en_cw; |
| 738 | penv_87->en_sw = penv_xmm->en_sw; |
| 739 | penv_87->en_tw = penv_xmm->en_tw; |
| 740 | penv_87->en_fip = penv_xmm->en_fip; |
| 741 | penv_87->en_fcs = penv_xmm->en_fcs; |
| 742 | penv_87->en_opcode = penv_xmm->en_opcode; |
| 743 | penv_87->en_foo = penv_xmm->en_foo; |
| 744 | penv_87->en_fos = penv_xmm->en_fos; |
| 745 | |
| 746 | /* FPU registers */ |
| 747 | for (i = 0; i < 8; ++i) |
| 748 | sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc; |
| 749 | } |
| 750 | |
| 751 | static void |
| 752 | set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm) |
| 753 | { |
| 754 | struct env87 *penv_87 = &sv_87->sv_env; |
| 755 | struct envxmm *penv_xmm = &sv_xmm->sv_env; |
| 756 | int i; |
| 757 | |
| 758 | /* FPU control/status */ |
| 759 | penv_xmm->en_cw = penv_87->en_cw; |
| 760 | penv_xmm->en_sw = penv_87->en_sw; |
| 761 | penv_xmm->en_tw = penv_87->en_tw; |
| 762 | penv_xmm->en_fip = penv_87->en_fip; |
| 763 | penv_xmm->en_fcs = penv_87->en_fcs; |
| 764 | penv_xmm->en_opcode = penv_87->en_opcode; |
| 765 | penv_xmm->en_foo = penv_87->en_foo; |
| 766 | penv_xmm->en_fos = penv_87->en_fos; |
| 767 | |
| 768 | /* FPU registers */ |
| 769 | for (i = 0; i < 8; ++i) |
| 770 | sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i]; |
| 771 | } |
| 772 | |
| 773 | int |
| 774 | fill_fpregs(struct lwp *lp, struct fpreg *fpregs) |
| 775 | { |
| 776 | if (lp->lwp_thread == NULL || lp->lwp_thread->td_pcb == NULL) |
| 777 | return EINVAL; |
| 778 | if (cpu_fxsr) { |
| 779 | fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm, |
| 780 | (struct save87 *)fpregs); |
| 781 | return (0); |
| 782 | } |
| 783 | bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs); |
| 784 | return (0); |
| 785 | } |
| 786 | |
| 787 | int |
| 788 | set_fpregs(struct lwp *lp, struct fpreg *fpregs) |
| 789 | { |
| 790 | if (cpu_fxsr) { |
| 791 | set_fpregs_xmm((struct save87 *)fpregs, |
| 792 | &lp->lwp_thread->td_pcb->pcb_save.sv_xmm); |
| 793 | return (0); |
| 794 | } |
| 795 | bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs); |
| 796 | return (0); |
| 797 | } |
| 798 | |
| 799 | int |
| 800 | fill_dbregs(struct lwp *lp, struct dbreg *dbregs) |
| 801 | { |
| 802 | return (ENOSYS); |
| 803 | } |
| 804 | |
| 805 | int |
| 806 | set_dbregs(struct lwp *lp, struct dbreg *dbregs) |
| 807 | { |
| 808 | return (ENOSYS); |
| 809 | } |
| 810 | |
| 811 | #if 0 |
| 812 | /* |
| 813 | * Return > 0 if a hardware breakpoint has been hit, and the |
| 814 | * breakpoint was in user space. Return 0, otherwise. |
| 815 | */ |
| 816 | int |
| 817 | user_dbreg_trap(void) |
| 818 | { |
| 819 | u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */ |
| 820 | u_int32_t bp; /* breakpoint bits extracted from dr6 */ |
| 821 | int nbp; /* number of breakpoints that triggered */ |
| 822 | caddr_t addr[4]; /* breakpoint addresses */ |
| 823 | int i; |
| 824 | |
| 825 | dr7 = rdr7(); |
| 826 | if ((dr7 & 0x000000ff) == 0) { |
| 827 | /* |
| 828 | * all GE and LE bits in the dr7 register are zero, |
| 829 | * thus the trap couldn't have been caused by the |
| 830 | * hardware debug registers |
| 831 | */ |
| 832 | return 0; |
| 833 | } |
| 834 | |
| 835 | nbp = 0; |
| 836 | dr6 = rdr6(); |
| 837 | bp = dr6 & 0x0000000f; |
| 838 | |
| 839 | if (!bp) { |
| 840 | /* |
| 841 | * None of the breakpoint bits are set meaning this |
| 842 | * trap was not caused by any of the debug registers |
| 843 | */ |
| 844 | return 0; |
| 845 | } |
| 846 | |
| 847 | /* |
| 848 | * at least one of the breakpoints were hit, check to see |
| 849 | * which ones and if any of them are user space addresses |
| 850 | */ |
| 851 | |
| 852 | if (bp & 0x01) { |
| 853 | addr[nbp++] = (caddr_t)rdr0(); |
| 854 | } |
| 855 | if (bp & 0x02) { |
| 856 | addr[nbp++] = (caddr_t)rdr1(); |
| 857 | } |
| 858 | if (bp & 0x04) { |
| 859 | addr[nbp++] = (caddr_t)rdr2(); |
| 860 | } |
| 861 | if (bp & 0x08) { |
| 862 | addr[nbp++] = (caddr_t)rdr3(); |
| 863 | } |
| 864 | |
| 865 | for (i=0; i<nbp; i++) { |
| 866 | if (addr[i] < |
| 867 | (caddr_t)VM_MAX_USER_ADDRESS) { |
| 868 | /* |
| 869 | * addr[i] is in user space |
| 870 | */ |
| 871 | return nbp; |
| 872 | } |
| 873 | } |
| 874 | |
| 875 | /* |
| 876 | * None of the breakpoints are in user space. |
| 877 | */ |
| 878 | return 0; |
| 879 | } |
| 880 | |
| 881 | #endif |
| 882 | |
| 883 | void |
| 884 | identcpu(void) |
| 885 | { |
| 886 | int regs[4]; |
| 887 | |
| 888 | do_cpuid(1, regs); |
| 889 | cpu_feature = regs[3]; |
| 890 | } |
| 891 | |
| 892 | |
| 893 | #ifndef DDB |
| 894 | void |
| 895 | Debugger(const char *msg) |
| 896 | { |
| 897 | kprintf("Debugger(\"%s\") called.\n", msg); |
| 898 | } |
| 899 | #endif /* no DDB */ |