15b81b6b3SRodney W. Grimes /*- 25b81b6b3SRodney W. Grimes * Copyright (c) 1982, 1986 The Regents of the University of California. 35b81b6b3SRodney W. Grimes * Copyright (c) 1989, 1990 William Jolitz 41561d038SDavid Greenman * Copyright (c) 1994 John Dyson 55b81b6b3SRodney W. Grimes * All rights reserved. 65b81b6b3SRodney W. Grimes * 75b81b6b3SRodney W. Grimes * This code is derived from software contributed to Berkeley by 85b81b6b3SRodney W. Grimes * the Systems Programming Group of the University of Utah Computer 95b81b6b3SRodney W. Grimes * Science Department, and William Jolitz. 105b81b6b3SRodney W. Grimes * 115b81b6b3SRodney W. Grimes * Redistribution and use in source and binary forms, with or without 125b81b6b3SRodney W. Grimes * modification, are permitted provided that the following conditions 135b81b6b3SRodney W. Grimes * are met: 145b81b6b3SRodney W. Grimes * 1. Redistributions of source code must retain the above copyright 155b81b6b3SRodney W. Grimes * notice, this list of conditions and the following disclaimer. 165b81b6b3SRodney W. Grimes * 2. Redistributions in binary form must reproduce the above copyright 175b81b6b3SRodney W. Grimes * notice, this list of conditions and the following disclaimer in the 185b81b6b3SRodney W. Grimes * documentation and/or other materials provided with the distribution. 195b81b6b3SRodney W. Grimes * 3. All advertising materials mentioning features or use of this software 205b81b6b3SRodney W. Grimes * must display the following acknowledgement: 215b81b6b3SRodney W. Grimes * This product includes software developed by the University of 225b81b6b3SRodney W. Grimes * California, Berkeley and its contributors. 235b81b6b3SRodney W. Grimes * 4. Neither the name of the University nor the names of its contributors 245b81b6b3SRodney W. Grimes * may be used to endorse or promote products derived from this software 255b81b6b3SRodney W. Grimes * without specific prior written permission. 265b81b6b3SRodney W. Grimes * 275b81b6b3SRodney W. Grimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 285b81b6b3SRodney W. Grimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 295b81b6b3SRodney W. Grimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 305b81b6b3SRodney W. Grimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 315b81b6b3SRodney W. Grimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 325b81b6b3SRodney W. Grimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 335b81b6b3SRodney W. Grimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 345b81b6b3SRodney W. Grimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 355b81b6b3SRodney W. Grimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 365b81b6b3SRodney W. Grimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 375b81b6b3SRodney W. Grimes * SUCH DAMAGE. 385b81b6b3SRodney W. Grimes * 39960173b9SRodney W. Grimes * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 405b81b6b3SRodney W. Grimes * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 415b81b6b3SRodney W. Grimes */ 425b81b6b3SRodney W. Grimes 439676a785SDavid E. O'Brien #include <sys/cdefs.h> 449676a785SDavid E. O'Brien __FBSDID("$FreeBSD$"); 459676a785SDavid E. O'Brien 466d7d4649SBruce Evans #include "opt_isa.h" 47558226eaSPeter Wemm #include "opt_npx.h" 4861324207SKATO Takenori #ifdef PC98 4961324207SKATO Takenori #include "opt_pc98.h" 5061324207SKATO Takenori #endif 510d74cc48SPeter Wemm #include "opt_reset.h" 52883bd55aSPoul-Henning Kamp #include "opt_cpu.h" 538890984dSGarrett Wollman 5426f9a767SRodney W. Grimes #include <sys/param.h> 5526f9a767SRodney W. Grimes #include <sys/systm.h> 569626b608SPoul-Henning Kamp #include <sys/bio.h> 5726f9a767SRodney W. Grimes #include <sys/buf.h> 586d7d4649SBruce Evans #include <sys/kse.h> 5966095752SJohn Dyson #include <sys/kernel.h> 600384fff8SJason Evans #include <sys/ktr.h> 616d7d4649SBruce Evans #include <sys/lock.h> 626d7d4649SBruce Evans #include <sys/malloc.h> 63411d10a6SAlan Cox #include <sys/mbuf.h> 6413b0500fSJohn Baldwin #include <sys/mutex.h> 655ad5504cSKelly Yancey #include <sys/pioctl.h> 666d7d4649SBruce Evans #include <sys/proc.h> 67e45db9b8SAlan Cox #include <sys/sf_buf.h> 686d7d4649SBruce Evans #include <sys/smp.h> 6966095752SJohn Dyson #include <sys/sysctl.h> 7091c28bfdSLuoqi Chen #include <sys/unistd.h> 716d7d4649SBruce Evans #include <sys/vnode.h> 726d7d4649SBruce Evans #include <sys/vmmeter.h> 735b81b6b3SRodney W. Grimes 74a2a1c95cSPeter Wemm #include <machine/cpu.h> 7541ee9f1cSPoul-Henning Kamp #include <machine/cputypes.h> 761f8745a9SPeter Wemm #include <machine/md_var.h> 7791c28bfdSLuoqi Chen #include <machine/pcb.h> 7891c28bfdSLuoqi Chen #include <machine/pcb_ext.h> 7991c28bfdSLuoqi Chen #include <machine/vm86.h> 805b81b6b3SRodney W. Grimes 81883bd55aSPoul-Henning Kamp #ifdef CPU_ELAN 82883bd55aSPoul-Henning Kamp #include <machine/elan_mmcr.h> 83883bd55aSPoul-Henning Kamp #endif 84883bd55aSPoul-Henning Kamp 8526f9a767SRodney W. Grimes #include <vm/vm.h> 866d7d4649SBruce Evans #include <vm/vm_extern.h> 8726f9a767SRodney W. Grimes #include <vm/vm_kern.h> 8824a1cce3SDavid Greenman #include <vm/vm_page.h> 89efeaf95aSDavid Greenman #include <vm/vm_map.h> 906d7d4649SBruce Evans #include <vm/vm_param.h> 915b81b6b3SRodney W. Grimes 92e30f0011SSatoshi Asami #ifdef PC98 93e30f0011SSatoshi Asami #include <pc98/pc98/pc98.h> 94e30f0011SSatoshi Asami #else 952320728fSRodney W. Grimes #include <i386/isa/isa.h> 96e30f0011SSatoshi Asami #endif 972320728fSRodney W. Grimes 98099a0e58SBosko Milekic #ifndef NSFBUFS 99099a0e58SBosko Milekic #define NSFBUFS (512 + maxusers * 16) 100099a0e58SBosko Milekic #endif 101099a0e58SBosko Milekic 10215fe3067SAlfred Perlstein static void cpu_reset_real(void); 1034260b00aSNate Lawson #ifdef SMP 1044260b00aSNate Lawson static void cpu_reset_proxy(void); 1054260b00aSNate Lawson static u_int cpu_reset_proxyid; 1064260b00aSNate Lawson static volatile u_int cpu_reset_proxy_active; 1074260b00aSNate Lawson #endif 108411d10a6SAlan Cox static void sf_buf_init(void *arg); 109411d10a6SAlan Cox SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL) 110411d10a6SAlan Cox 1110543fa53SAlan Cox LIST_HEAD(sf_head, sf_buf); 112411d10a6SAlan Cox 1130543fa53SAlan Cox /* 1140543fa53SAlan Cox * A hash table of active sendfile(2) buffers 1150543fa53SAlan Cox */ 1160543fa53SAlan Cox static struct sf_head *sf_buf_active; 1170543fa53SAlan Cox static u_long sf_buf_hashmask; 1180543fa53SAlan Cox 1190543fa53SAlan Cox #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask) 1200543fa53SAlan Cox 121a5819cb5SAlan Cox static TAILQ_HEAD(, sf_buf) sf_buf_freelist; 122411d10a6SAlan Cox static u_int sf_buf_alloc_want; 123411d10a6SAlan Cox 1240543fa53SAlan Cox /* 1250543fa53SAlan Cox * A lock used to synchronize access to the hash table and free list 1260543fa53SAlan Cox */ 1270543fa53SAlan Cox static struct mtx sf_buf_lock; 1280543fa53SAlan Cox 12937b087a6SPeter Wemm extern int _ucodesel, _udatasel; 1302f1e7069STor Egge 131a4f7a4c9SDavid Greenman /* 1325b81b6b3SRodney W. Grimes * Finish a fork operation, with process p2 nearly set up. 133a2a1c95cSPeter Wemm * Copy and update the pcb, set up the stack so that the child 134a2a1c95cSPeter Wemm * ready to run and return to user mode. 1355b81b6b3SRodney W. Grimes */ 136a2a1c95cSPeter Wemm void 137079b7badSJulian Elischer cpu_fork(td1, p2, td2, flags) 138b40ce416SJulian Elischer register struct thread *td1; 139b40ce416SJulian Elischer register struct proc *p2; 140079b7badSJulian Elischer struct thread *td2; 14191c28bfdSLuoqi Chen int flags; 1425b81b6b3SRodney W. Grimes { 143b40ce416SJulian Elischer register struct proc *p1; 14491c28bfdSLuoqi Chen struct pcb *pcb2; 14524db0459SJohn Baldwin struct mdproc *mdp2; 146a4b8c657SBruce Evans #ifdef DEV_NPX 147ba74981eSWarner Losh register_t savecrit; 148a4b8c657SBruce Evans #endif 14991c28bfdSLuoqi Chen 150b40ce416SJulian Elischer p1 = td1->td_proc; 15191c28bfdSLuoqi Chen if ((flags & RFPROC) == 0) { 15291c28bfdSLuoqi Chen if ((flags & RFMEM) == 0) { 15391c28bfdSLuoqi Chen /* unshare user LDT */ 15445449f84SJulian Elischer struct mdproc *mdp1 = &p1->p_md; 15524db0459SJohn Baldwin struct proc_ldt *pldt = mdp1->md_ldt; 15624db0459SJohn Baldwin if (pldt && pldt->ldt_refcnt > 1) { 15724db0459SJohn Baldwin pldt = user_ldt_alloc(mdp1, pldt->ldt_len); 15824db0459SJohn Baldwin if (pldt == NULL) 1591acf256dSJohn Baldwin panic("could not copy LDT"); 16024db0459SJohn Baldwin mdp1->md_ldt = pldt; 16124db0459SJohn Baldwin set_user_ldt(mdp1); 16224db0459SJohn Baldwin user_ldt_free(td1); 16391c28bfdSLuoqi Chen } 16491c28bfdSLuoqi Chen } 16591c28bfdSLuoqi Chen return; 16691c28bfdSLuoqi Chen } 1675b81b6b3SRodney W. Grimes 1681f8745a9SPeter Wemm /* Ensure that p1's pcb is up to date. */ 169a4b8c657SBruce Evans #ifdef DEV_NPX 170b40ce416SJulian Elischer if (td1 == curthread) 171b40ce416SJulian Elischer td1->td_pcb->pcb_gs = rgs(); 172ba74981eSWarner Losh savecrit = intr_disable(); 1730bbc8826SJohn Baldwin if (PCPU_GET(fpcurthread) == td1) 174b40ce416SJulian Elischer npxsave(&td1->td_pcb->pcb_save); 175ba74981eSWarner Losh intr_restore(savecrit); 1769f449d2aSBruce Evans #endif 1771f8745a9SPeter Wemm 178b40ce416SJulian Elischer /* Point the pcb to the top of the stack */ 1796041f1a5SBill Paul pcb2 = (struct pcb *)(td2->td_kstack + 1806041f1a5SBill Paul td2->td_kstack_pages * PAGE_SIZE) - 1; 181b40ce416SJulian Elischer td2->td_pcb = pcb2; 182b40ce416SJulian Elischer 18324db0459SJohn Baldwin /* Copy p1's pcb */ 184b40ce416SJulian Elischer bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 185a2a1c95cSPeter Wemm 18624db0459SJohn Baldwin /* Point mdproc and then copy over td1's contents */ 18745449f84SJulian Elischer mdp2 = &p2->p_md; 18845449f84SJulian Elischer bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 18924db0459SJohn Baldwin 190a2a1c95cSPeter Wemm /* 191a2a1c95cSPeter Wemm * Create a new fresh stack for the new process. 1921f8745a9SPeter Wemm * Copy the trap frame for the return to user mode as if from a 193a4b8c657SBruce Evans * syscall. This copies most of the user mode register values. 194e5a860ebSPeter Wemm * The -16 is so we can expand the trapframe if we go to vm86. 195a2a1c95cSPeter Wemm */ 196e5a860ebSPeter Wemm td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1; 197b40ce416SJulian Elischer bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 198a2a1c95cSPeter Wemm 199b40ce416SJulian Elischer td2->td_frame->tf_eax = 0; /* Child returns zero */ 200b40ce416SJulian Elischer td2->td_frame->tf_eflags &= ~PSL_C; /* success */ 201b40ce416SJulian Elischer td2->td_frame->tf_edx = 1; 2020d54b9d1SJohn Baldwin 203a2a1c95cSPeter Wemm /* 2045ad5504cSKelly Yancey * If the parent process has the trap bit set (i.e. a debugger had 2055ad5504cSKelly Yancey * single stepped the process to the system call), we need to clear 2065ad5504cSKelly Yancey * the trap flag from the new frame unless the debugger had set PF_FORK 2075ad5504cSKelly Yancey * on the parent. Otherwise, the child will receive a (likely 2085ad5504cSKelly Yancey * unexpected) SIGTRAP when it executes the first instruction after 2095ad5504cSKelly Yancey * returning to userland. 2105ad5504cSKelly Yancey */ 2115ad5504cSKelly Yancey if ((p1->p_pfsflags & PF_FORK) == 0) 2125ad5504cSKelly Yancey td2->td_frame->tf_eflags &= ~PSL_T; 2135ad5504cSKelly Yancey 2145ad5504cSKelly Yancey /* 215a2a1c95cSPeter Wemm * Set registers for trampoline to user mode. Leave space for the 216a2a1c95cSPeter Wemm * return address on stack. These are the kernel mode register values. 217a2a1c95cSPeter Wemm */ 2187ab9b220SJake Burkholder #ifdef PAE 2197ab9b220SJake Burkholder pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt); 2207ab9b220SJake Burkholder #else 221b1028ad1SLuoqi Chen pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 2227ab9b220SJake Burkholder #endif 22337b087a6SPeter Wemm pcb2->pcb_edi = 0; 22437b087a6SPeter Wemm pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 22537b087a6SPeter Wemm pcb2->pcb_ebp = 0; 226b40ce416SJulian Elischer pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); 227b40ce416SJulian Elischer pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */ 2281f8745a9SPeter Wemm pcb2->pcb_eip = (int)fork_trampoline; 229a93b6bf5SThomas Moestl pcb2->pcb_psl = PSL_KERNEL; /* ints disabled */ 2302257a44fSDavid Xu pcb2->pcb_gs = rgs(); 231a4b8c657SBruce Evans /*- 232a4b8c657SBruce Evans * pcb2->pcb_dr*: cloned above. 2331f8745a9SPeter Wemm * pcb2->pcb_savefpu: cloned above. 234a4b8c657SBruce Evans * pcb2->pcb_flags: cloned above. 2351f8745a9SPeter Wemm * pcb2->pcb_onfault: cloned above (always NULL here?). 236a4b8c657SBruce Evans * pcb2->pcb_gs: cloned above. 237a4b8c657SBruce Evans * pcb2->pcb_ext: cleared below. 2381f8745a9SPeter Wemm */ 2395b81b6b3SRodney W. Grimes 24048a09cf2SJohn Dyson /* 24148a09cf2SJohn Dyson * XXX don't copy the i/o pages. this should probably be fixed. 24248a09cf2SJohn Dyson */ 24348a09cf2SJohn Dyson pcb2->pcb_ext = 0; 24448a09cf2SJohn Dyson 2458c39a127SStefan Eßer /* Copy the LDT, if necessary. */ 246df4d012bSJohn Baldwin mtx_lock_spin(&sched_lock); 247753d1af1SJohn Baldwin if (mdp2->md_ldt != NULL) { 24891c28bfdSLuoqi Chen if (flags & RFMEM) { 24924db0459SJohn Baldwin mdp2->md_ldt->ldt_refcnt++; 25091c28bfdSLuoqi Chen } else { 25124db0459SJohn Baldwin mdp2->md_ldt = user_ldt_alloc(mdp2, 25224db0459SJohn Baldwin mdp2->md_ldt->ldt_len); 25324db0459SJohn Baldwin if (mdp2->md_ldt == NULL) 254df4d012bSJohn Baldwin panic("could not copy LDT"); 25591c28bfdSLuoqi Chen } 2568c39a127SStefan Eßer } 257df4d012bSJohn Baldwin mtx_unlock_spin(&sched_lock); 2588c39a127SStefan Eßer 259a2a1c95cSPeter Wemm /* 260a2a1c95cSPeter Wemm * Now, cpu_switch() can schedule the new process. 261a2a1c95cSPeter Wemm * pcb_esp is loaded pointing to the cpu_switch() stack frame 262a2a1c95cSPeter Wemm * containing the return address when exiting cpu_switch. 26337b087a6SPeter Wemm * This will normally be to fork_trampoline(), which will have 26437b087a6SPeter Wemm * %ebx loaded with the new proc's pointer. fork_trampoline() 265a2a1c95cSPeter Wemm * will set up a stack to call fork_return(p, frame); to complete 266a2a1c95cSPeter Wemm * the return to user-mode. 267a2a1c95cSPeter Wemm */ 268a2a1c95cSPeter Wemm } 269a2a1c95cSPeter Wemm 270a2a1c95cSPeter Wemm /* 271a2a1c95cSPeter Wemm * Intercept the return address from a freshly forked process that has NOT 272a2a1c95cSPeter Wemm * been scheduled yet. 273a2a1c95cSPeter Wemm * 274a2a1c95cSPeter Wemm * This is needed to make kernel threads stay in kernel mode. 275a2a1c95cSPeter Wemm */ 276a2a1c95cSPeter Wemm void 277b40ce416SJulian Elischer cpu_set_fork_handler(td, func, arg) 278b40ce416SJulian Elischer struct thread *td; 27915fe3067SAlfred Perlstein void (*func)(void *); 2809c8b8baaSPeter Wemm void *arg; 281a2a1c95cSPeter Wemm { 282a2a1c95cSPeter Wemm /* 283a2a1c95cSPeter Wemm * Note that the trap frame follows the args, so the function 284a2a1c95cSPeter Wemm * is really called like this: func(arg, frame); 285a2a1c95cSPeter Wemm */ 286b40ce416SJulian Elischer td->td_pcb->pcb_esi = (int) func; /* function */ 287b40ce416SJulian Elischer td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 2885b81b6b3SRodney W. Grimes } 2895b81b6b3SRodney W. Grimes 2907c2b54e8SNate Williams void 291e602ba25SJulian Elischer cpu_exit(struct thread *td) 292e602ba25SJulian Elischer { 293e602ba25SJulian Elischer struct mdproc *mdp; 294e602ba25SJulian Elischer 29566250360SDavid Xu /* Reset pc->pcb_gs and %gs before possibly invalidating it. */ 296e602ba25SJulian Elischer mdp = &td->td_proc->p_md; 29766250360SDavid Xu if (mdp->md_ldt) { 29866250360SDavid Xu td->td_pcb->pcb_gs = _udatasel; 29966250360SDavid Xu load_gs(_udatasel); 300e602ba25SJulian Elischer user_ldt_free(td); 30166250360SDavid Xu } 302e602ba25SJulian Elischer } 303e602ba25SJulian Elischer 304e602ba25SJulian Elischer void 305e602ba25SJulian Elischer cpu_thread_exit(struct thread *td) 3065b81b6b3SRodney W. Grimes { 307b40ce416SJulian Elischer struct pcb *pcb = td->td_pcb; 3084cc99cf6SJohn Baldwin 309558226eaSPeter Wemm #ifdef DEV_NPX 310d5e1f581SDavid Xu if (td == PCPU_GET(fpcurthread)) 311d5e1f581SDavid Xu npxdrop(); 312558226eaSPeter Wemm #endif 313b19d9defSMaxime Henrion if (pcb->pcb_flags & PCB_DBREGS) { 3146745e78cSBrian S. Dean /* disable all hardware breakpoints */ 315b19d9defSMaxime Henrion reset_dbregs(); 316b19d9defSMaxime Henrion pcb->pcb_flags &= ~PCB_DBREGS; 317b19d9defSMaxime Henrion } 318b19d9defSMaxime Henrion } 319b19d9defSMaxime Henrion 320b19d9defSMaxime Henrion void 321696058c3SJulian Elischer cpu_thread_clean(struct thread *td) 322b19d9defSMaxime Henrion { 323b19d9defSMaxime Henrion struct pcb *pcb; 324b19d9defSMaxime Henrion 325b19d9defSMaxime Henrion pcb = td->td_pcb; 32648a09cf2SJohn Dyson if (pcb->pcb_ext != 0) { 327e602ba25SJulian Elischer /* XXXKSE XXXSMP not SMP SAFE.. what locks do we have? */ 328e602ba25SJulian Elischer /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 32948a09cf2SJohn Dyson /* 33048a09cf2SJohn Dyson * XXX do we need to move the TSS off the allocated pages 33148a09cf2SJohn Dyson * before freeing them? (not done here) 33248a09cf2SJohn Dyson */ 33348a09cf2SJohn Dyson kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 33448a09cf2SJohn Dyson ctob(IOPAGES + 1)); 33548a09cf2SJohn Dyson pcb->pcb_ext = 0; 33648a09cf2SJohn Dyson } 3375b81b6b3SRodney W. Grimes } 3385b81b6b3SRodney W. Grimes 339381fe1aaSGarrett Wollman void 340710338e9SMarcel Moolenaar cpu_thread_swapin(struct thread *td) 341710338e9SMarcel Moolenaar { 342710338e9SMarcel Moolenaar } 343710338e9SMarcel Moolenaar 344710338e9SMarcel Moolenaar void 345710338e9SMarcel Moolenaar cpu_thread_swapout(struct thread *td) 346710338e9SMarcel Moolenaar { 347710338e9SMarcel Moolenaar } 348710338e9SMarcel Moolenaar 349710338e9SMarcel Moolenaar void 350e602ba25SJulian Elischer cpu_thread_setup(struct thread *td) 351e602ba25SJulian Elischer { 352e602ba25SJulian Elischer 3536041f1a5SBill Paul td->td_pcb = (struct pcb *)(td->td_kstack + 3546041f1a5SBill Paul td->td_kstack_pages * PAGE_SIZE) - 1; 355e602ba25SJulian Elischer td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1; 356c7251aedSTor Egge td->td_pcb->pcb_ext = NULL; 357e602ba25SJulian Elischer } 358e602ba25SJulian Elischer 359e602ba25SJulian Elischer /* 360575525a0SJonathan Mini * Initialize machine state (pcb and trap frame) for a new thread about to 36148bfcdddSJulian Elischer * upcall. Pu t enough state in the new thread's PCB to get it to go back 36248bfcdddSJulian Elischer * userret(), where we can intercept it again to set the return (upcall) 36348bfcdddSJulian Elischer * Address and stack, along with those from upcals that are from other sources 36448bfcdddSJulian Elischer * such as those generated in thread_userret() itself. 365e602ba25SJulian Elischer */ 366e602ba25SJulian Elischer void 36711e0f8e1SMarcel Moolenaar cpu_set_upcall(struct thread *td, struct thread *td0) 368e602ba25SJulian Elischer { 369e602ba25SJulian Elischer struct pcb *pcb2; 370e602ba25SJulian Elischer 371e602ba25SJulian Elischer /* Point the pcb to the top of the stack. */ 372e602ba25SJulian Elischer pcb2 = td->td_pcb; 373e602ba25SJulian Elischer 374e602ba25SJulian Elischer /* 375e602ba25SJulian Elischer * Copy the upcall pcb. This loads kernel regs. 376e602ba25SJulian Elischer * Those not loaded individually below get their default 377e602ba25SJulian Elischer * values here. 378e602ba25SJulian Elischer * 379e602ba25SJulian Elischer * XXXKSE It might be a good idea to simply skip this as 380e602ba25SJulian Elischer * the values of the other registers may be unimportant. 381e602ba25SJulian Elischer * This would remove any requirement for knowing the KSE 382e602ba25SJulian Elischer * at this time (see the matching comment below for 383e602ba25SJulian Elischer * more analysis) (need a good safe default). 384e602ba25SJulian Elischer */ 38511e0f8e1SMarcel Moolenaar bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 386d5e1f581SDavid Xu pcb2->pcb_flags &= ~(PCB_NPXTRAP|PCB_NPXINITDONE); 387e602ba25SJulian Elischer 388e602ba25SJulian Elischer /* 389e602ba25SJulian Elischer * Create a new fresh stack for the new thread. 390e602ba25SJulian Elischer * The -16 is so we can expand the trapframe if we go to vm86. 391e602ba25SJulian Elischer * Don't forget to set this stack value into whatever supplies 392e602ba25SJulian Elischer * the address for the fault handlers. 393e602ba25SJulian Elischer * The contexts are filled in at the time we actually DO the 394e602ba25SJulian Elischer * upcall as only then do we know which KSE we got. 395e602ba25SJulian Elischer */ 39611e0f8e1SMarcel Moolenaar bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 397e602ba25SJulian Elischer 398e602ba25SJulian Elischer /* 399e602ba25SJulian Elischer * Set registers for trampoline to user mode. Leave space for the 400e602ba25SJulian Elischer * return address on stack. These are the kernel mode register values. 401e602ba25SJulian Elischer */ 4027ab9b220SJake Burkholder #ifdef PAE 4037ab9b220SJake Burkholder pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdpt); 4047ab9b220SJake Burkholder #else 405e602ba25SJulian Elischer pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir); 4067ab9b220SJake Burkholder #endif 407e602ba25SJulian Elischer pcb2->pcb_edi = 0; 408e602ba25SJulian Elischer pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 409e602ba25SJulian Elischer pcb2->pcb_ebp = 0; 410e602ba25SJulian Elischer pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */ 411e602ba25SJulian Elischer pcb2->pcb_ebx = (int)td; /* trampoline arg */ 412e602ba25SJulian Elischer pcb2->pcb_eip = (int)fork_trampoline; 413e602ba25SJulian Elischer pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */ 4142257a44fSDavid Xu pcb2->pcb_gs = rgs(); 415e602ba25SJulian Elischer /* 416e602ba25SJulian Elischer * If we didn't copy the pcb, we'd need to do the following registers: 417e602ba25SJulian Elischer * pcb2->pcb_dr*: cloned above. 418e602ba25SJulian Elischer * pcb2->pcb_savefpu: cloned above. 419e602ba25SJulian Elischer * pcb2->pcb_flags: cloned above. 420e602ba25SJulian Elischer * pcb2->pcb_onfault: cloned above (always NULL here?). 421e602ba25SJulian Elischer * pcb2->pcb_gs: cloned above. XXXKSE ??? 422e602ba25SJulian Elischer * pcb2->pcb_ext: cleared below. 423e602ba25SJulian Elischer */ 424e602ba25SJulian Elischer pcb2->pcb_ext = NULL; 425e602ba25SJulian Elischer } 426e602ba25SJulian Elischer 427e602ba25SJulian Elischer /* 42848bfcdddSJulian Elischer * Set that machine state for performing an upcall that has to 42948bfcdddSJulian Elischer * be done in thread_userret() so that those upcalls generated 43048bfcdddSJulian Elischer * in thread_userret() itself can be done as well. 431e602ba25SJulian Elischer */ 432575525a0SJonathan Mini void 4335215b187SJeff Roberson cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 434575525a0SJonathan Mini { 435575525a0SJonathan Mini 436575525a0SJonathan Mini /* 437696058c3SJulian Elischer * Do any extra cleaning that needs to be done. 438696058c3SJulian Elischer * The thread may have optional components 439696058c3SJulian Elischer * that are not present in a fresh thread. 440696058c3SJulian Elischer * This may be a recycled thread so make it look 441696058c3SJulian Elischer * as though it's newly allocated. 442696058c3SJulian Elischer */ 443696058c3SJulian Elischer cpu_thread_clean(td); 444696058c3SJulian Elischer 445696058c3SJulian Elischer /* 446575525a0SJonathan Mini * Set the trap frame to point at the beginning of the uts 447575525a0SJonathan Mini * function. 448575525a0SJonathan Mini */ 4492396628bSDavid Xu td->td_frame->tf_ebp = 0; 450575525a0SJonathan Mini td->td_frame->tf_esp = 4515215b187SJeff Roberson (int)ku->ku_stack.ss_sp + ku->ku_stack.ss_size - 16; 4525215b187SJeff Roberson td->td_frame->tf_eip = (int)ku->ku_func; 453575525a0SJonathan Mini 454575525a0SJonathan Mini /* 455575525a0SJonathan Mini * Pass the address of the mailbox for this kse to the uts 456575525a0SJonathan Mini * function as a parameter on the stack. 457575525a0SJonathan Mini */ 458575525a0SJonathan Mini suword((void *)(td->td_frame->tf_esp + sizeof(void *)), 4595215b187SJeff Roberson (int)ku->ku_mailbox); 460e602ba25SJulian Elischer } 461e602ba25SJulian Elischer 4625b81b6b3SRodney W. Grimes /* 4635b81b6b3SRodney W. Grimes * Convert kernel VA to physical address 4645b81b6b3SRodney W. Grimes */ 465227f9a1cSJake Burkholder vm_paddr_t 4667f8cb368SDavid Greenman kvtop(void *addr) 4675b81b6b3SRodney W. Grimes { 468227f9a1cSJake Burkholder vm_paddr_t pa; 4695b81b6b3SRodney W. Grimes 470227f9a1cSJake Burkholder pa = pmap_kextract((vm_offset_t)addr); 471227f9a1cSJake Burkholder if (pa == 0) 4725b81b6b3SRodney W. Grimes panic("kvtop: zero page frame"); 473227f9a1cSJake Burkholder return (pa); 4745b81b6b3SRodney W. Grimes } 4755b81b6b3SRodney W. Grimes 4764260b00aSNate Lawson #ifdef SMP 4774260b00aSNate Lawson static void 4784260b00aSNate Lawson cpu_reset_proxy() 4794260b00aSNate Lawson { 4804260b00aSNate Lawson 4814260b00aSNate Lawson cpu_reset_proxy_active = 1; 4824260b00aSNate Lawson while (cpu_reset_proxy_active == 1) 4834260b00aSNate Lawson ; /* Wait for other cpu to see that we've started */ 4844260b00aSNate Lawson stop_cpus((1<<cpu_reset_proxyid)); 4854260b00aSNate Lawson printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 4864260b00aSNate Lawson DELAY(1000000); 4874260b00aSNate Lawson cpu_reset_real(); 4884260b00aSNate Lawson } 4894260b00aSNate Lawson #endif 4904260b00aSNate Lawson 4917f8cb368SDavid Greenman void 492d447dbeeSBruce Evans cpu_reset() 493d447dbeeSBruce Evans { 4942f1e7069STor Egge #ifdef SMP 4954260b00aSNate Lawson u_int cnt, map; 4962f1e7069STor Egge 49763a6daf6SNate Lawson if (smp_active) { 498ef73ae4bSJake Burkholder map = PCPU_GET(other_cpus) & ~stopped_cpus; 4992f1e7069STor Egge if (map != 0) { 5002f1e7069STor Egge printf("cpu_reset: Stopping other CPUs\n"); 50163a6daf6SNate Lawson stop_cpus(map); 5022f1e7069STor Egge } 5034260b00aSNate Lawson 5044260b00aSNate Lawson if (PCPU_GET(cpuid) != 0) { 5054260b00aSNate Lawson cpu_reset_proxyid = PCPU_GET(cpuid); 5064260b00aSNate Lawson cpustop_restartfunc = cpu_reset_proxy; 5074260b00aSNate Lawson cpu_reset_proxy_active = 0; 5084260b00aSNate Lawson printf("cpu_reset: Restarting BSP\n"); 5094260b00aSNate Lawson started_cpus = (1<<0); /* Restart CPU #0 */ 5104260b00aSNate Lawson 5114260b00aSNate Lawson cnt = 0; 5124260b00aSNate Lawson while (cpu_reset_proxy_active == 0 && cnt < 10000000) 5134260b00aSNate Lawson cnt++; /* Wait for BSP to announce restart */ 5144260b00aSNate Lawson if (cpu_reset_proxy_active == 0) 5154260b00aSNate Lawson printf("cpu_reset: Failed to restart BSP\n"); 5164260b00aSNate Lawson enable_intr(); 5174260b00aSNate Lawson cpu_reset_proxy_active = 2; 5184260b00aSNate Lawson 5194260b00aSNate Lawson while (1); 5204260b00aSNate Lawson /* NOTREACHED */ 5214260b00aSNate Lawson } 5224260b00aSNate Lawson 5232f1e7069STor Egge DELAY(1000000); 5242f1e7069STor Egge } 5252f1e7069STor Egge #endif 52663a6daf6SNate Lawson cpu_reset_real(); 52763a6daf6SNate Lawson /* NOTREACHED */ 5282f1e7069STor Egge } 5292f1e7069STor Egge 5302f1e7069STor Egge static void 5312f1e7069STor Egge cpu_reset_real() 5322f1e7069STor Egge { 533d447dbeeSBruce Evans 534883bd55aSPoul-Henning Kamp #ifdef CPU_ELAN 535883bd55aSPoul-Henning Kamp if (elan_mmcr != NULL) 536883bd55aSPoul-Henning Kamp elan_mmcr->RESCFG = 1; 537883bd55aSPoul-Henning Kamp #endif 538883bd55aSPoul-Henning Kamp 53902d12d93SPoul-Henning Kamp if (cpu == CPU_GEODE1100) { 54002d12d93SPoul-Henning Kamp /* Attempt Geode's own reset */ 54102d12d93SPoul-Henning Kamp outl(0xcf8, 0x80009044ul); 54202d12d93SPoul-Henning Kamp outl(0xcfc, 0xf); 54302d12d93SPoul-Henning Kamp } 54402d12d93SPoul-Henning Kamp 54503245f09SKATO Takenori #ifdef PC98 54603245f09SKATO Takenori /* 54703245f09SKATO Takenori * Attempt to do a CPU reset via CPU reset port. 54803245f09SKATO Takenori */ 549d447dbeeSBruce Evans disable_intr(); 5508e929d65SKATO Takenori if ((inb(0x35) & 0xa0) != 0xa0) { 55103245f09SKATO Takenori outb(0x37, 0x0f); /* SHUT0 = 0. */ 55203245f09SKATO Takenori outb(0x37, 0x0b); /* SHUT1 = 0. */ 5538e929d65SKATO Takenori } 55403245f09SKATO Takenori outb(0xf0, 0x00); /* Reset. */ 55503245f09SKATO Takenori #else 55663a6daf6SNate Lawson #if !defined(BROKEN_KEYBOARD_RESET) 5572320728fSRodney W. Grimes /* 5582320728fSRodney W. Grimes * Attempt to do a CPU reset via the keyboard controller, 55963a6daf6SNate Lawson * do not turn off GateA20, as any machine that fails 5602320728fSRodney W. Grimes * to do the reset here would then end up in no man's land. 5612320728fSRodney W. Grimes */ 5622320728fSRodney W. Grimes outb(IO_KBD + 4, 0xFE); 5632320728fSRodney W. Grimes DELAY(500000); /* wait 0.5 sec to see if that did it */ 5642320728fSRodney W. Grimes printf("Keyboard reset did not work, attempting CPU shutdown\n"); 5652320728fSRodney W. Grimes DELAY(1000000); /* wait 1 sec for printf to complete */ 5665eb46edfSDavid Greenman #endif 56703245f09SKATO Takenori #endif /* PC98 */ 56863a6daf6SNate Lawson 56963a6daf6SNate Lawson /* Force a shutdown by unmapping entire address space. */ 570910548deSJake Burkholder bzero((caddr_t)PTD, NBPTD); 5715b81b6b3SRodney W. Grimes 5725b81b6b3SRodney W. Grimes /* "good night, sweet prince .... <THUNK!>" */ 57327e9b35eSJohn Dyson invltlb(); 5745b81b6b3SRodney W. Grimes /* NOTREACHED */ 5757f8cb368SDavid Greenman while(1); 5765b81b6b3SRodney W. Grimes } 577b9d60b3fSDavid Greenman 578e0b78e19SJoerg Wunsch /* 579411d10a6SAlan Cox * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-)) 580411d10a6SAlan Cox */ 581411d10a6SAlan Cox static void 582411d10a6SAlan Cox sf_buf_init(void *arg) 583411d10a6SAlan Cox { 584411d10a6SAlan Cox struct sf_buf *sf_bufs; 585411d10a6SAlan Cox vm_offset_t sf_base; 586411d10a6SAlan Cox int i; 587411d10a6SAlan Cox 588099a0e58SBosko Milekic nsfbufs = NSFBUFS; 589099a0e58SBosko Milekic TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs); 590099a0e58SBosko Milekic 5910543fa53SAlan Cox sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask); 592a5819cb5SAlan Cox TAILQ_INIT(&sf_buf_freelist); 593411d10a6SAlan Cox sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE); 594411d10a6SAlan Cox sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, 595411d10a6SAlan Cox M_NOWAIT | M_ZERO); 596411d10a6SAlan Cox for (i = 0; i < nsfbufs; i++) { 597411d10a6SAlan Cox sf_bufs[i].kva = sf_base + i * PAGE_SIZE; 598a5819cb5SAlan Cox TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry); 599411d10a6SAlan Cox } 600411d10a6SAlan Cox sf_buf_alloc_want = 0; 6010543fa53SAlan Cox mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF); 602411d10a6SAlan Cox } 603411d10a6SAlan Cox 604411d10a6SAlan Cox /* 605411d10a6SAlan Cox * Get an sf_buf from the freelist. Will block if none are available. 606411d10a6SAlan Cox */ 607411d10a6SAlan Cox struct sf_buf * 608d3cb0d99SAlan Cox sf_buf_alloc(struct vm_page *m, int flags) 609411d10a6SAlan Cox { 61020351fafSAlan Cox pt_entry_t opte, *ptep; 6110543fa53SAlan Cox struct sf_head *hash_list; 612411d10a6SAlan Cox struct sf_buf *sf; 613411d10a6SAlan Cox int error; 614411d10a6SAlan Cox 6150543fa53SAlan Cox hash_list = &sf_buf_active[SF_BUF_HASH(m)]; 6160543fa53SAlan Cox mtx_lock(&sf_buf_lock); 6170543fa53SAlan Cox LIST_FOREACH(sf, hash_list, list_entry) { 6180543fa53SAlan Cox if (sf->m == m) { 6190543fa53SAlan Cox sf->ref_count++; 6205eda9873SMike Silbersack if (sf->ref_count == 1) { 6219a63fc0dSAlan Cox TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 6225eda9873SMike Silbersack nsfbufsused++; 6235caf2b00SMike Silbersack nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 6245eda9873SMike Silbersack } 6250543fa53SAlan Cox goto done; 6260543fa53SAlan Cox } 6270543fa53SAlan Cox } 628a5819cb5SAlan Cox while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) { 629d3cb0d99SAlan Cox if (flags & SFB_NOWAIT) 630d3cb0d99SAlan Cox goto done; 631411d10a6SAlan Cox sf_buf_alloc_want++; 632ddeb5b24SMike Silbersack mbstat.sf_allocwait++; 633d3cb0d99SAlan Cox error = msleep(&sf_buf_freelist, &sf_buf_lock, 634d3cb0d99SAlan Cox (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0); 635411d10a6SAlan Cox sf_buf_alloc_want--; 636411d10a6SAlan Cox 637411d10a6SAlan Cox /* 638411d10a6SAlan Cox * If we got a signal, don't risk going back to sleep. 639411d10a6SAlan Cox */ 640411d10a6SAlan Cox if (error) 6410543fa53SAlan Cox goto done; 642411d10a6SAlan Cox } 643a5819cb5SAlan Cox TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 644a5819cb5SAlan Cox if (sf->m != NULL) 6450543fa53SAlan Cox LIST_REMOVE(sf, list_entry); 6460543fa53SAlan Cox LIST_INSERT_HEAD(hash_list, sf, list_entry); 6470543fa53SAlan Cox sf->ref_count = 1; 648411d10a6SAlan Cox sf->m = m; 6495eda9873SMike Silbersack nsfbufsused++; 6505caf2b00SMike Silbersack nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 65120351fafSAlan Cox 65220351fafSAlan Cox /* 65320351fafSAlan Cox * Update the sf_buf's virtual-to-physical mapping, flushing the 65420351fafSAlan Cox * virtual address from the TLB only if the PTE implies that the old 65520351fafSAlan Cox * mapping has been used. Since the reference count for the sf_buf's 65620351fafSAlan Cox * old mapping was zero, that mapping is not currently in use. 65720351fafSAlan Cox * Consequently, there is no need to exchange the old and new PTEs 65820351fafSAlan Cox * atomically, even under PAE. 65920351fafSAlan Cox */ 66020351fafSAlan Cox ptep = vtopte(sf->kva); 66120351fafSAlan Cox opte = *ptep; 66220351fafSAlan Cox *ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V; 66320351fafSAlan Cox if ((opte & (PG_A | PG_V)) == (PG_A | PG_V)) 66420351fafSAlan Cox pmap_invalidate_page(kernel_pmap, sf->kva); 6650543fa53SAlan Cox done: 6660543fa53SAlan Cox mtx_unlock(&sf_buf_lock); 667411d10a6SAlan Cox return (sf); 668411d10a6SAlan Cox } 669411d10a6SAlan Cox 670411d10a6SAlan Cox /* 67190ecfebdSAlan Cox * Remove a reference from the given sf_buf, adding it to the free 67290ecfebdSAlan Cox * list when its reference count reaches zero. A freed sf_buf still, 67390ecfebdSAlan Cox * however, retains its virtual-to-physical mapping until it is 67490ecfebdSAlan Cox * recycled or reactivated by sf_buf_alloc(9). 675411d10a6SAlan Cox */ 676411d10a6SAlan Cox void 67790ecfebdSAlan Cox sf_buf_free(struct sf_buf *sf) 678411d10a6SAlan Cox { 679411d10a6SAlan Cox 6800543fa53SAlan Cox mtx_lock(&sf_buf_lock); 6810543fa53SAlan Cox sf->ref_count--; 6820543fa53SAlan Cox if (sf->ref_count == 0) { 683a5819cb5SAlan Cox TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry); 68490ecfebdSAlan Cox nsfbufsused--; 6850543fa53SAlan Cox if (sf_buf_alloc_want > 0) 6860543fa53SAlan Cox wakeup_one(&sf_buf_freelist); 6870543fa53SAlan Cox } 6880543fa53SAlan Cox mtx_unlock(&sf_buf_lock); 689411d10a6SAlan Cox } 690411d10a6SAlan Cox 691411d10a6SAlan Cox /* 69257d7d7b3SJustin T. Gibbs * Software interrupt handler for queued VM system processing. 69357d7d7b3SJustin T. Gibbs */ 69457d7d7b3SJustin T. Gibbs void 6958088699fSJohn Baldwin swi_vm(void *dummy) 69657d7d7b3SJustin T. Gibbs { 69757d7d7b3SJustin T. Gibbs if (busdma_swi_pending != 0) 69857d7d7b3SJustin T. Gibbs busdma_swi(); 69957d7d7b3SJustin T. Gibbs } 70057d7d7b3SJustin T. Gibbs 70157d7d7b3SJustin T. Gibbs /* 702cae6f73aSJoerg Wunsch * Tell whether this address is in some physical memory region. 703e0b78e19SJoerg Wunsch * Currently used by the kernel coredump code in order to avoid 704e0b78e19SJoerg Wunsch * dumping the ``ISA memory hole'' which could cause indefinite hangs, 705e0b78e19SJoerg Wunsch * or other unpredictable behaviour. 706e0b78e19SJoerg Wunsch */ 707e0b78e19SJoerg Wunsch 708e0b78e19SJoerg Wunsch int 7092c38d78eSAlan Cox is_physical_memory(vm_paddr_t addr) 710e0b78e19SJoerg Wunsch { 711e0b78e19SJoerg Wunsch 71203927d3cSPeter Wemm #ifdef DEV_ISA 713e0b78e19SJoerg Wunsch /* The ISA ``memory hole''. */ 714e0b78e19SJoerg Wunsch if (addr >= 0xa0000 && addr < 0x100000) 715cae6f73aSJoerg Wunsch return 0; 716e0b78e19SJoerg Wunsch #endif 717e0b78e19SJoerg Wunsch 718e0b78e19SJoerg Wunsch /* 719e0b78e19SJoerg Wunsch * stuff other tests for known memory-mapped devices (PCI?) 720e0b78e19SJoerg Wunsch * here 721e0b78e19SJoerg Wunsch */ 722e0b78e19SJoerg Wunsch 723cae6f73aSJoerg Wunsch return 1; 724e0b78e19SJoerg Wunsch } 725