1dfe7b3bfSKonstantin Belousov /*- 2dfe7b3bfSKonstantin Belousov * Copyright (c) 2003 Peter Wemm. 3dfe7b3bfSKonstantin Belousov * Copyright (c) 1992 Terrence R. Lambert. 4dfe7b3bfSKonstantin Belousov * Copyright (c) 1982, 1987, 1990 The Regents of the University of California. 5dfe7b3bfSKonstantin Belousov * All rights reserved. 6dfe7b3bfSKonstantin Belousov * 7dfe7b3bfSKonstantin Belousov * This code is derived from software contributed to Berkeley by 8dfe7b3bfSKonstantin Belousov * William Jolitz. 9dfe7b3bfSKonstantin Belousov * 10dfe7b3bfSKonstantin Belousov * Redistribution and use in source and binary forms, with or without 11dfe7b3bfSKonstantin Belousov * modification, are permitted provided that the following conditions 12dfe7b3bfSKonstantin Belousov * are met: 13dfe7b3bfSKonstantin Belousov * 1. Redistributions of source code must retain the above copyright 14dfe7b3bfSKonstantin Belousov * notice, this list of conditions and the following disclaimer. 15dfe7b3bfSKonstantin Belousov * 2. Redistributions in binary form must reproduce the above copyright 16dfe7b3bfSKonstantin Belousov * notice, this list of conditions and the following disclaimer in the 17dfe7b3bfSKonstantin Belousov * documentation and/or other materials provided with the distribution. 18dfe7b3bfSKonstantin Belousov * 3. All advertising materials mentioning features or use of this software 19dfe7b3bfSKonstantin Belousov * must display the following acknowledgement: 20dfe7b3bfSKonstantin Belousov * This product includes software developed by the University of 21dfe7b3bfSKonstantin Belousov * California, Berkeley and its contributors. 22dfe7b3bfSKonstantin Belousov * 4. Neither the name of the University nor the names of its contributors 23dfe7b3bfSKonstantin Belousov * may be used to endorse or promote products derived from this software 24dfe7b3bfSKonstantin Belousov * without specific prior written permission. 25dfe7b3bfSKonstantin Belousov * 26dfe7b3bfSKonstantin Belousov * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27dfe7b3bfSKonstantin Belousov * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28dfe7b3bfSKonstantin Belousov * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29dfe7b3bfSKonstantin Belousov * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30dfe7b3bfSKonstantin Belousov * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31dfe7b3bfSKonstantin Belousov * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32dfe7b3bfSKonstantin Belousov * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33dfe7b3bfSKonstantin Belousov * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34dfe7b3bfSKonstantin Belousov * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35dfe7b3bfSKonstantin Belousov * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36dfe7b3bfSKonstantin Belousov * SUCH DAMAGE. 37dfe7b3bfSKonstantin Belousov * 38dfe7b3bfSKonstantin Belousov * from: @(#)machdep.c 7.4 (Berkeley) 6/3/91 39dfe7b3bfSKonstantin Belousov */ 40dfe7b3bfSKonstantin Belousov 41dfe7b3bfSKonstantin Belousov #include <sys/cdefs.h> 42dfe7b3bfSKonstantin Belousov __FBSDID("$FreeBSD$"); 43dfe7b3bfSKonstantin Belousov 44dfe7b3bfSKonstantin Belousov #include "opt_atpic.h" 45dfe7b3bfSKonstantin Belousov #include "opt_compat.h" 46dfe7b3bfSKonstantin Belousov #include "opt_cpu.h" 47dfe7b3bfSKonstantin Belousov #include "opt_ddb.h" 48dfe7b3bfSKonstantin Belousov #include "opt_inet.h" 49dfe7b3bfSKonstantin Belousov #include "opt_isa.h" 50835c2787SKonstantin Belousov #include "opt_kdb.h" 51dfe7b3bfSKonstantin Belousov #include "opt_kstack_pages.h" 52dfe7b3bfSKonstantin Belousov #include "opt_maxmem.h" 53dfe7b3bfSKonstantin Belousov #include "opt_mp_watchdog.h" 54dfe7b3bfSKonstantin Belousov #include "opt_platform.h" 55dfe7b3bfSKonstantin Belousov #ifdef __i386__ 56dfe7b3bfSKonstantin Belousov #include "opt_apic.h" 57dfe7b3bfSKonstantin Belousov #endif 58dfe7b3bfSKonstantin Belousov 59dfe7b3bfSKonstantin Belousov #include <sys/param.h> 60dfe7b3bfSKonstantin Belousov #include <sys/proc.h> 61dfe7b3bfSKonstantin Belousov #include <sys/systm.h> 62dfe7b3bfSKonstantin Belousov #include <sys/bus.h> 63dfe7b3bfSKonstantin Belousov #include <sys/cpu.h> 64dfe7b3bfSKonstantin Belousov #include <sys/kdb.h> 65dfe7b3bfSKonstantin Belousov #include <sys/kernel.h> 66dfe7b3bfSKonstantin Belousov #include <sys/ktr.h> 67dfe7b3bfSKonstantin Belousov #include <sys/lock.h> 68dfe7b3bfSKonstantin Belousov #include <sys/malloc.h> 69dfe7b3bfSKonstantin Belousov #include <sys/mutex.h> 70dfe7b3bfSKonstantin Belousov #include <sys/pcpu.h> 71dfe7b3bfSKonstantin Belousov #include <sys/rwlock.h> 72dfe7b3bfSKonstantin Belousov #include <sys/sched.h> 73dfe7b3bfSKonstantin Belousov #ifdef SMP 74dfe7b3bfSKonstantin Belousov #include <sys/smp.h> 75dfe7b3bfSKonstantin Belousov #endif 76dfe7b3bfSKonstantin Belousov #include <sys/sysctl.h> 77dfe7b3bfSKonstantin Belousov 78dfe7b3bfSKonstantin Belousov #include <machine/clock.h> 79dfe7b3bfSKonstantin Belousov #include <machine/cpu.h> 80dfe7b3bfSKonstantin Belousov #include <machine/cputypes.h> 81dfe7b3bfSKonstantin Belousov #include <machine/specialreg.h> 82dfe7b3bfSKonstantin Belousov #include <machine/md_var.h> 83dfe7b3bfSKonstantin Belousov #include <machine/mp_watchdog.h> 84dfe7b3bfSKonstantin Belousov #include <machine/tss.h> 85dfe7b3bfSKonstantin Belousov #ifdef SMP 86dfe7b3bfSKonstantin Belousov #include <machine/smp.h> 87dfe7b3bfSKonstantin Belousov #endif 88b57a73f8SKonstantin Belousov #include <x86/acpica_machdep.h> 89dfe7b3bfSKonstantin Belousov 90dfe7b3bfSKonstantin Belousov #include <vm/vm.h> 91dfe7b3bfSKonstantin Belousov #include <vm/vm_extern.h> 92dfe7b3bfSKonstantin Belousov #include <vm/vm_kern.h> 93dfe7b3bfSKonstantin Belousov #include <vm/vm_page.h> 94dfe7b3bfSKonstantin Belousov #include <vm/vm_map.h> 95dfe7b3bfSKonstantin Belousov #include <vm/vm_object.h> 96dfe7b3bfSKonstantin Belousov #include <vm/vm_pager.h> 97dfe7b3bfSKonstantin Belousov #include <vm/vm_param.h> 98dfe7b3bfSKonstantin Belousov 998428d0f1SAndriy Gapon #include <isa/isareg.h> 1008428d0f1SAndriy Gapon 101d9e8bbb6SKonstantin Belousov #define STATE_RUNNING 0x0 102d9e8bbb6SKonstantin Belousov #define STATE_MWAIT 0x1 103d9e8bbb6SKonstantin Belousov #define STATE_SLEEPING 0x2 104d9e8bbb6SKonstantin Belousov 1058428d0f1SAndriy Gapon #ifdef SMP 1068428d0f1SAndriy Gapon static u_int cpu_reset_proxyid; 1078428d0f1SAndriy Gapon static volatile u_int cpu_reset_proxy_active; 1088428d0f1SAndriy Gapon #endif 1098428d0f1SAndriy Gapon 1108428d0f1SAndriy Gapon 111dfe7b3bfSKonstantin Belousov /* 112dfe7b3bfSKonstantin Belousov * Machine dependent boot() routine 113dfe7b3bfSKonstantin Belousov * 114dfe7b3bfSKonstantin Belousov * I haven't seen anything to put here yet 115dfe7b3bfSKonstantin Belousov * Possibly some stuff might be grafted back here from boot() 116dfe7b3bfSKonstantin Belousov */ 117dfe7b3bfSKonstantin Belousov void 118dfe7b3bfSKonstantin Belousov cpu_boot(int howto) 119dfe7b3bfSKonstantin Belousov { 120dfe7b3bfSKonstantin Belousov } 121dfe7b3bfSKonstantin Belousov 122dfe7b3bfSKonstantin Belousov /* 123dfe7b3bfSKonstantin Belousov * Flush the D-cache for non-DMA I/O so that the I-cache can 124dfe7b3bfSKonstantin Belousov * be made coherent later. 125dfe7b3bfSKonstantin Belousov */ 126dfe7b3bfSKonstantin Belousov void 127dfe7b3bfSKonstantin Belousov cpu_flush_dcache(void *ptr, size_t len) 128dfe7b3bfSKonstantin Belousov { 129dfe7b3bfSKonstantin Belousov /* Not applicable */ 130dfe7b3bfSKonstantin Belousov } 131dfe7b3bfSKonstantin Belousov 132b57a73f8SKonstantin Belousov void 133b57a73f8SKonstantin Belousov acpi_cpu_c1(void) 134b57a73f8SKonstantin Belousov { 135b57a73f8SKonstantin Belousov 136b57a73f8SKonstantin Belousov __asm __volatile("sti; hlt"); 137b57a73f8SKonstantin Belousov } 138b57a73f8SKonstantin Belousov 13919d4720bSJonathan T. Looney /* 14019d4720bSJonathan T. Looney * Use mwait to pause execution while waiting for an interrupt or 14119d4720bSJonathan T. Looney * another thread to signal that there is more work. 14219d4720bSJonathan T. Looney * 14319d4720bSJonathan T. Looney * NOTE: Interrupts will cause a wakeup; however, this function does 14419d4720bSJonathan T. Looney * not enable interrupt handling. The caller is responsible to enable 14519d4720bSJonathan T. Looney * interrupts. 14619d4720bSJonathan T. Looney */ 147b57a73f8SKonstantin Belousov void 148b57a73f8SKonstantin Belousov acpi_cpu_idle_mwait(uint32_t mwait_hint) 149b57a73f8SKonstantin Belousov { 150b57a73f8SKonstantin Belousov int *state; 151b57a73f8SKonstantin Belousov 152b57a73f8SKonstantin Belousov /* 153319117fdSKonstantin Belousov * A comment in Linux patch claims that 'CPUs run faster with 154319117fdSKonstantin Belousov * speculation protection disabled. All CPU threads in a core 155319117fdSKonstantin Belousov * must disable speculation protection for it to be 156319117fdSKonstantin Belousov * disabled. Disable it while we are idle so the other 157319117fdSKonstantin Belousov * hyperthread can run fast.' 158319117fdSKonstantin Belousov * 159b57a73f8SKonstantin Belousov * XXXKIB. Software coordination mode should be supported, 160b57a73f8SKonstantin Belousov * but all Intel CPUs provide hardware coordination. 161b57a73f8SKonstantin Belousov */ 162d9e8bbb6SKonstantin Belousov 163d9e8bbb6SKonstantin Belousov state = (int *)PCPU_PTR(monitorbuf); 164d9e8bbb6SKonstantin Belousov KASSERT(*state == STATE_SLEEPING, 165d9e8bbb6SKonstantin Belousov ("cpu_mwait_cx: wrong monitorbuf state")); 166d9e8bbb6SKonstantin Belousov *state = STATE_MWAIT; 167319117fdSKonstantin Belousov handle_ibrs_entry(); 168b57a73f8SKonstantin Belousov cpu_monitor(state, 0, 0); 169d9e8bbb6SKonstantin Belousov if (*state == STATE_MWAIT) 170b57a73f8SKonstantin Belousov cpu_mwait(MWAIT_INTRBREAK, mwait_hint); 171319117fdSKonstantin Belousov handle_ibrs_exit(); 172d9e8bbb6SKonstantin Belousov 173d9e8bbb6SKonstantin Belousov /* 174d9e8bbb6SKonstantin Belousov * We should exit on any event that interrupts mwait, because 175d9e8bbb6SKonstantin Belousov * that event might be a wanted interrupt. 176d9e8bbb6SKonstantin Belousov */ 177d9e8bbb6SKonstantin Belousov *state = STATE_RUNNING; 178b57a73f8SKonstantin Belousov } 179b57a73f8SKonstantin Belousov 180dfe7b3bfSKonstantin Belousov /* Get current clock frequency for the given cpu id. */ 181dfe7b3bfSKonstantin Belousov int 182dfe7b3bfSKonstantin Belousov cpu_est_clockrate(int cpu_id, uint64_t *rate) 183dfe7b3bfSKonstantin Belousov { 184dfe7b3bfSKonstantin Belousov uint64_t tsc1, tsc2; 185dfe7b3bfSKonstantin Belousov uint64_t acnt, mcnt, perf; 186dfe7b3bfSKonstantin Belousov register_t reg; 187dfe7b3bfSKonstantin Belousov 188dfe7b3bfSKonstantin Belousov if (pcpu_find(cpu_id) == NULL || rate == NULL) 189dfe7b3bfSKonstantin Belousov return (EINVAL); 190dfe7b3bfSKonstantin Belousov #ifdef __i386__ 191dfe7b3bfSKonstantin Belousov if ((cpu_feature & CPUID_TSC) == 0) 192dfe7b3bfSKonstantin Belousov return (EOPNOTSUPP); 193dfe7b3bfSKonstantin Belousov #endif 194dfe7b3bfSKonstantin Belousov 195dfe7b3bfSKonstantin Belousov /* 196dfe7b3bfSKonstantin Belousov * If TSC is P-state invariant and APERF/MPERF MSRs do not exist, 197dfe7b3bfSKonstantin Belousov * DELAY(9) based logic fails. 198dfe7b3bfSKonstantin Belousov */ 199dfe7b3bfSKonstantin Belousov if (tsc_is_invariant && !tsc_perf_stat) 200dfe7b3bfSKonstantin Belousov return (EOPNOTSUPP); 201dfe7b3bfSKonstantin Belousov 202dfe7b3bfSKonstantin Belousov #ifdef SMP 203dfe7b3bfSKonstantin Belousov if (smp_cpus > 1) { 204dfe7b3bfSKonstantin Belousov /* Schedule ourselves on the indicated cpu. */ 205dfe7b3bfSKonstantin Belousov thread_lock(curthread); 206dfe7b3bfSKonstantin Belousov sched_bind(curthread, cpu_id); 207dfe7b3bfSKonstantin Belousov thread_unlock(curthread); 208dfe7b3bfSKonstantin Belousov } 209dfe7b3bfSKonstantin Belousov #endif 210dfe7b3bfSKonstantin Belousov 211dfe7b3bfSKonstantin Belousov /* Calibrate by measuring a short delay. */ 212dfe7b3bfSKonstantin Belousov reg = intr_disable(); 213dfe7b3bfSKonstantin Belousov if (tsc_is_invariant) { 214dfe7b3bfSKonstantin Belousov wrmsr(MSR_MPERF, 0); 215dfe7b3bfSKonstantin Belousov wrmsr(MSR_APERF, 0); 216dfe7b3bfSKonstantin Belousov tsc1 = rdtsc(); 217dfe7b3bfSKonstantin Belousov DELAY(1000); 218dfe7b3bfSKonstantin Belousov mcnt = rdmsr(MSR_MPERF); 219dfe7b3bfSKonstantin Belousov acnt = rdmsr(MSR_APERF); 220dfe7b3bfSKonstantin Belousov tsc2 = rdtsc(); 221dfe7b3bfSKonstantin Belousov intr_restore(reg); 222dfe7b3bfSKonstantin Belousov perf = 1000 * acnt / mcnt; 223dfe7b3bfSKonstantin Belousov *rate = (tsc2 - tsc1) * perf; 224dfe7b3bfSKonstantin Belousov } else { 225dfe7b3bfSKonstantin Belousov tsc1 = rdtsc(); 226dfe7b3bfSKonstantin Belousov DELAY(1000); 227dfe7b3bfSKonstantin Belousov tsc2 = rdtsc(); 228dfe7b3bfSKonstantin Belousov intr_restore(reg); 229dfe7b3bfSKonstantin Belousov *rate = (tsc2 - tsc1) * 1000; 230dfe7b3bfSKonstantin Belousov } 231dfe7b3bfSKonstantin Belousov 232dfe7b3bfSKonstantin Belousov #ifdef SMP 233dfe7b3bfSKonstantin Belousov if (smp_cpus > 1) { 234dfe7b3bfSKonstantin Belousov thread_lock(curthread); 235dfe7b3bfSKonstantin Belousov sched_unbind(curthread); 236dfe7b3bfSKonstantin Belousov thread_unlock(curthread); 237dfe7b3bfSKonstantin Belousov } 238dfe7b3bfSKonstantin Belousov #endif 239dfe7b3bfSKonstantin Belousov 240dfe7b3bfSKonstantin Belousov return (0); 241dfe7b3bfSKonstantin Belousov } 242dfe7b3bfSKonstantin Belousov 243dfe7b3bfSKonstantin Belousov /* 244dfe7b3bfSKonstantin Belousov * Shutdown the CPU as much as possible 245dfe7b3bfSKonstantin Belousov */ 246dfe7b3bfSKonstantin Belousov void 247dfe7b3bfSKonstantin Belousov cpu_halt(void) 248dfe7b3bfSKonstantin Belousov { 249dfe7b3bfSKonstantin Belousov for (;;) 250dfe7b3bfSKonstantin Belousov halt(); 251dfe7b3bfSKonstantin Belousov } 252dfe7b3bfSKonstantin Belousov 2538428d0f1SAndriy Gapon static void 254*b7b25af0SAndriy Gapon cpu_reset_real(void) 2558428d0f1SAndriy Gapon { 2568428d0f1SAndriy Gapon struct region_descriptor null_idt; 2578428d0f1SAndriy Gapon int b; 2588428d0f1SAndriy Gapon 2598428d0f1SAndriy Gapon disable_intr(); 2608428d0f1SAndriy Gapon #ifdef CPU_ELAN 2618428d0f1SAndriy Gapon if (elan_mmcr != NULL) 2628428d0f1SAndriy Gapon elan_mmcr->RESCFG = 1; 2638428d0f1SAndriy Gapon #endif 2648428d0f1SAndriy Gapon #ifdef __i386__ 2658428d0f1SAndriy Gapon if (cpu == CPU_GEODE1100) { 2668428d0f1SAndriy Gapon /* Attempt Geode's own reset */ 2678428d0f1SAndriy Gapon outl(0xcf8, 0x80009044ul); 2688428d0f1SAndriy Gapon outl(0xcfc, 0xf); 2698428d0f1SAndriy Gapon } 2708428d0f1SAndriy Gapon #endif 2718428d0f1SAndriy Gapon #if !defined(BROKEN_KEYBOARD_RESET) 2728428d0f1SAndriy Gapon /* 2738428d0f1SAndriy Gapon * Attempt to do a CPU reset via the keyboard controller, 2748428d0f1SAndriy Gapon * do not turn off GateA20, as any machine that fails 2758428d0f1SAndriy Gapon * to do the reset here would then end up in no man's land. 2768428d0f1SAndriy Gapon */ 2778428d0f1SAndriy Gapon outb(IO_KBD + 4, 0xFE); 2788428d0f1SAndriy Gapon DELAY(500000); /* wait 0.5 sec to see if that did it */ 2798428d0f1SAndriy Gapon #endif 2808428d0f1SAndriy Gapon 2818428d0f1SAndriy Gapon /* 2828428d0f1SAndriy Gapon * Attempt to force a reset via the Reset Control register at 2838428d0f1SAndriy Gapon * I/O port 0xcf9. Bit 2 forces a system reset when it 2848428d0f1SAndriy Gapon * transitions from 0 to 1. Bit 1 selects the type of reset 2858428d0f1SAndriy Gapon * to attempt: 0 selects a "soft" reset, and 1 selects a 2868428d0f1SAndriy Gapon * "hard" reset. We try a "hard" reset. The first write sets 2878428d0f1SAndriy Gapon * bit 1 to select a "hard" reset and clears bit 2. The 2888428d0f1SAndriy Gapon * second write forces a 0 -> 1 transition in bit 2 to trigger 2898428d0f1SAndriy Gapon * a reset. 2908428d0f1SAndriy Gapon */ 2918428d0f1SAndriy Gapon outb(0xcf9, 0x2); 2928428d0f1SAndriy Gapon outb(0xcf9, 0x6); 2938428d0f1SAndriy Gapon DELAY(500000); /* wait 0.5 sec to see if that did it */ 2948428d0f1SAndriy Gapon 2958428d0f1SAndriy Gapon /* 2968428d0f1SAndriy Gapon * Attempt to force a reset via the Fast A20 and Init register 2978428d0f1SAndriy Gapon * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 2988428d0f1SAndriy Gapon * Bit 0 asserts INIT# when set to 1. We are careful to only 2998428d0f1SAndriy Gapon * preserve bit 1 while setting bit 0. We also must clear bit 3008428d0f1SAndriy Gapon * 0 before setting it if it isn't already clear. 3018428d0f1SAndriy Gapon */ 3028428d0f1SAndriy Gapon b = inb(0x92); 3038428d0f1SAndriy Gapon if (b != 0xff) { 3048428d0f1SAndriy Gapon if ((b & 0x1) != 0) 3058428d0f1SAndriy Gapon outb(0x92, b & 0xfe); 3068428d0f1SAndriy Gapon outb(0x92, b | 0x1); 3078428d0f1SAndriy Gapon DELAY(500000); /* wait 0.5 sec to see if that did it */ 3088428d0f1SAndriy Gapon } 3098428d0f1SAndriy Gapon 3108428d0f1SAndriy Gapon printf("No known reset method worked, attempting CPU shutdown\n"); 3118428d0f1SAndriy Gapon DELAY(1000000); /* wait 1 sec for printf to complete */ 3128428d0f1SAndriy Gapon 3138428d0f1SAndriy Gapon /* Wipe the IDT. */ 3148428d0f1SAndriy Gapon null_idt.rd_limit = 0; 3158428d0f1SAndriy Gapon null_idt.rd_base = 0; 3168428d0f1SAndriy Gapon lidt(&null_idt); 3178428d0f1SAndriy Gapon 3188428d0f1SAndriy Gapon /* "good night, sweet prince .... <THUNK!>" */ 3198428d0f1SAndriy Gapon breakpoint(); 3208428d0f1SAndriy Gapon 3218428d0f1SAndriy Gapon /* NOTREACHED */ 3228428d0f1SAndriy Gapon while(1); 3238428d0f1SAndriy Gapon } 3248428d0f1SAndriy Gapon 3258428d0f1SAndriy Gapon #ifdef SMP 3268428d0f1SAndriy Gapon static void 327*b7b25af0SAndriy Gapon cpu_reset_proxy(void) 3288428d0f1SAndriy Gapon { 3298428d0f1SAndriy Gapon 3308428d0f1SAndriy Gapon cpu_reset_proxy_active = 1; 3318428d0f1SAndriy Gapon while (cpu_reset_proxy_active == 1) 3328428d0f1SAndriy Gapon ia32_pause(); /* Wait for other cpu to see that we've started */ 3338428d0f1SAndriy Gapon 3348428d0f1SAndriy Gapon printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 3358428d0f1SAndriy Gapon DELAY(1000000); 3368428d0f1SAndriy Gapon cpu_reset_real(); 3378428d0f1SAndriy Gapon } 3388428d0f1SAndriy Gapon #endif 3398428d0f1SAndriy Gapon 3408428d0f1SAndriy Gapon void 341*b7b25af0SAndriy Gapon cpu_reset(void) 3428428d0f1SAndriy Gapon { 3438428d0f1SAndriy Gapon #ifdef SMP 3448428d0f1SAndriy Gapon cpuset_t map; 3458428d0f1SAndriy Gapon u_int cnt; 3468428d0f1SAndriy Gapon 3478428d0f1SAndriy Gapon if (smp_started) { 3488428d0f1SAndriy Gapon map = all_cpus; 3498428d0f1SAndriy Gapon CPU_CLR(PCPU_GET(cpuid), &map); 3508428d0f1SAndriy Gapon CPU_NAND(&map, &stopped_cpus); 3518428d0f1SAndriy Gapon if (!CPU_EMPTY(&map)) { 3528428d0f1SAndriy Gapon printf("cpu_reset: Stopping other CPUs\n"); 3538428d0f1SAndriy Gapon stop_cpus(map); 3548428d0f1SAndriy Gapon } 3558428d0f1SAndriy Gapon 3568428d0f1SAndriy Gapon if (PCPU_GET(cpuid) != 0) { 3578428d0f1SAndriy Gapon cpu_reset_proxyid = PCPU_GET(cpuid); 3588428d0f1SAndriy Gapon cpustop_restartfunc = cpu_reset_proxy; 3598428d0f1SAndriy Gapon cpu_reset_proxy_active = 0; 3608428d0f1SAndriy Gapon printf("cpu_reset: Restarting BSP\n"); 3618428d0f1SAndriy Gapon 3628428d0f1SAndriy Gapon /* Restart CPU #0. */ 3638428d0f1SAndriy Gapon CPU_SETOF(0, &started_cpus); 3648428d0f1SAndriy Gapon wmb(); 3658428d0f1SAndriy Gapon 3668428d0f1SAndriy Gapon cnt = 0; 3678428d0f1SAndriy Gapon while (cpu_reset_proxy_active == 0 && cnt < 10000000) { 3688428d0f1SAndriy Gapon ia32_pause(); 3698428d0f1SAndriy Gapon cnt++; /* Wait for BSP to announce restart */ 3708428d0f1SAndriy Gapon } 3718428d0f1SAndriy Gapon if (cpu_reset_proxy_active == 0) { 3728428d0f1SAndriy Gapon printf("cpu_reset: Failed to restart BSP\n"); 3738428d0f1SAndriy Gapon } else { 3748428d0f1SAndriy Gapon cpu_reset_proxy_active = 2; 3758428d0f1SAndriy Gapon while (1) 3768428d0f1SAndriy Gapon ia32_pause(); 3778428d0f1SAndriy Gapon /* NOTREACHED */ 3788428d0f1SAndriy Gapon } 3798428d0f1SAndriy Gapon } 3808428d0f1SAndriy Gapon 3818428d0f1SAndriy Gapon DELAY(1000000); 3828428d0f1SAndriy Gapon } 3838428d0f1SAndriy Gapon #endif 3848428d0f1SAndriy Gapon cpu_reset_real(); 3858428d0f1SAndriy Gapon /* NOTREACHED */ 3868428d0f1SAndriy Gapon } 3878428d0f1SAndriy Gapon 388b57a73f8SKonstantin Belousov bool 389b57a73f8SKonstantin Belousov cpu_mwait_usable(void) 390b57a73f8SKonstantin Belousov { 391b57a73f8SKonstantin Belousov 392b57a73f8SKonstantin Belousov return ((cpu_feature2 & CPUID2_MON) != 0 && ((cpu_mon_mwait_flags & 393b57a73f8SKonstantin Belousov (CPUID5_MON_MWAIT_EXT | CPUID5_MWAIT_INTRBREAK)) == 394b57a73f8SKonstantin Belousov (CPUID5_MON_MWAIT_EXT | CPUID5_MWAIT_INTRBREAK))); 395b57a73f8SKonstantin Belousov } 396b57a73f8SKonstantin Belousov 397dfe7b3bfSKonstantin Belousov void (*cpu_idle_hook)(sbintime_t) = NULL; /* ACPI idle hook. */ 398dfe7b3bfSKonstantin Belousov static int cpu_ident_amdc1e = 0; /* AMD C1E supported. */ 399dfe7b3bfSKonstantin Belousov static int idle_mwait = 1; /* Use MONITOR/MWAIT for short idle. */ 400dfe7b3bfSKonstantin Belousov SYSCTL_INT(_machdep, OID_AUTO, idle_mwait, CTLFLAG_RWTUN, &idle_mwait, 401dfe7b3bfSKonstantin Belousov 0, "Use MONITOR/MWAIT for short idle"); 402dfe7b3bfSKonstantin Belousov 403dfe7b3bfSKonstantin Belousov static void 404dfe7b3bfSKonstantin Belousov cpu_idle_acpi(sbintime_t sbt) 405dfe7b3bfSKonstantin Belousov { 406dfe7b3bfSKonstantin Belousov int *state; 407dfe7b3bfSKonstantin Belousov 408dfe7b3bfSKonstantin Belousov state = (int *)PCPU_PTR(monitorbuf); 409dfe7b3bfSKonstantin Belousov *state = STATE_SLEEPING; 410dfe7b3bfSKonstantin Belousov 411dfe7b3bfSKonstantin Belousov /* See comments in cpu_idle_hlt(). */ 412dfe7b3bfSKonstantin Belousov disable_intr(); 413dfe7b3bfSKonstantin Belousov if (sched_runnable()) 414dfe7b3bfSKonstantin Belousov enable_intr(); 415dfe7b3bfSKonstantin Belousov else if (cpu_idle_hook) 416dfe7b3bfSKonstantin Belousov cpu_idle_hook(sbt); 417dfe7b3bfSKonstantin Belousov else 418b57a73f8SKonstantin Belousov acpi_cpu_c1(); 419dfe7b3bfSKonstantin Belousov *state = STATE_RUNNING; 420dfe7b3bfSKonstantin Belousov } 421dfe7b3bfSKonstantin Belousov 422dfe7b3bfSKonstantin Belousov static void 423dfe7b3bfSKonstantin Belousov cpu_idle_hlt(sbintime_t sbt) 424dfe7b3bfSKonstantin Belousov { 425dfe7b3bfSKonstantin Belousov int *state; 426dfe7b3bfSKonstantin Belousov 427dfe7b3bfSKonstantin Belousov state = (int *)PCPU_PTR(monitorbuf); 428dfe7b3bfSKonstantin Belousov *state = STATE_SLEEPING; 429dfe7b3bfSKonstantin Belousov 430dfe7b3bfSKonstantin Belousov /* 431dfe7b3bfSKonstantin Belousov * Since we may be in a critical section from cpu_idle(), if 432dfe7b3bfSKonstantin Belousov * an interrupt fires during that critical section we may have 433dfe7b3bfSKonstantin Belousov * a pending preemption. If the CPU halts, then that thread 434dfe7b3bfSKonstantin Belousov * may not execute until a later interrupt awakens the CPU. 435dfe7b3bfSKonstantin Belousov * To handle this race, check for a runnable thread after 436dfe7b3bfSKonstantin Belousov * disabling interrupts and immediately return if one is 437dfe7b3bfSKonstantin Belousov * found. Also, we must absolutely guarentee that hlt is 438dfe7b3bfSKonstantin Belousov * the next instruction after sti. This ensures that any 439dfe7b3bfSKonstantin Belousov * interrupt that fires after the call to disable_intr() will 440dfe7b3bfSKonstantin Belousov * immediately awaken the CPU from hlt. Finally, please note 441dfe7b3bfSKonstantin Belousov * that on x86 this works fine because of interrupts enabled only 442dfe7b3bfSKonstantin Belousov * after the instruction following sti takes place, while IF is set 443dfe7b3bfSKonstantin Belousov * to 1 immediately, allowing hlt instruction to acknowledge the 444dfe7b3bfSKonstantin Belousov * interrupt. 445dfe7b3bfSKonstantin Belousov */ 446dfe7b3bfSKonstantin Belousov disable_intr(); 447dfe7b3bfSKonstantin Belousov if (sched_runnable()) 448dfe7b3bfSKonstantin Belousov enable_intr(); 449dfe7b3bfSKonstantin Belousov else 450b57a73f8SKonstantin Belousov acpi_cpu_c1(); 451dfe7b3bfSKonstantin Belousov *state = STATE_RUNNING; 452dfe7b3bfSKonstantin Belousov } 453dfe7b3bfSKonstantin Belousov 454dfe7b3bfSKonstantin Belousov static void 455dfe7b3bfSKonstantin Belousov cpu_idle_mwait(sbintime_t sbt) 456dfe7b3bfSKonstantin Belousov { 457dfe7b3bfSKonstantin Belousov int *state; 458dfe7b3bfSKonstantin Belousov 459dfe7b3bfSKonstantin Belousov state = (int *)PCPU_PTR(monitorbuf); 460dfe7b3bfSKonstantin Belousov *state = STATE_MWAIT; 461dfe7b3bfSKonstantin Belousov 462dfe7b3bfSKonstantin Belousov /* See comments in cpu_idle_hlt(). */ 463dfe7b3bfSKonstantin Belousov disable_intr(); 464dfe7b3bfSKonstantin Belousov if (sched_runnable()) { 465dfe7b3bfSKonstantin Belousov enable_intr(); 466dfe7b3bfSKonstantin Belousov *state = STATE_RUNNING; 467dfe7b3bfSKonstantin Belousov return; 468dfe7b3bfSKonstantin Belousov } 469dfe7b3bfSKonstantin Belousov cpu_monitor(state, 0, 0); 470dfe7b3bfSKonstantin Belousov if (*state == STATE_MWAIT) 471dfe7b3bfSKonstantin Belousov __asm __volatile("sti; mwait" : : "a" (MWAIT_C1), "c" (0)); 472dfe7b3bfSKonstantin Belousov else 473dfe7b3bfSKonstantin Belousov enable_intr(); 474dfe7b3bfSKonstantin Belousov *state = STATE_RUNNING; 475dfe7b3bfSKonstantin Belousov } 476dfe7b3bfSKonstantin Belousov 477dfe7b3bfSKonstantin Belousov static void 478dfe7b3bfSKonstantin Belousov cpu_idle_spin(sbintime_t sbt) 479dfe7b3bfSKonstantin Belousov { 480dfe7b3bfSKonstantin Belousov int *state; 481dfe7b3bfSKonstantin Belousov int i; 482dfe7b3bfSKonstantin Belousov 483dfe7b3bfSKonstantin Belousov state = (int *)PCPU_PTR(monitorbuf); 484dfe7b3bfSKonstantin Belousov *state = STATE_RUNNING; 485dfe7b3bfSKonstantin Belousov 486dfe7b3bfSKonstantin Belousov /* 487dfe7b3bfSKonstantin Belousov * The sched_runnable() call is racy but as long as there is 488dfe7b3bfSKonstantin Belousov * a loop missing it one time will have just a little impact if any 489dfe7b3bfSKonstantin Belousov * (and it is much better than missing the check at all). 490dfe7b3bfSKonstantin Belousov */ 491dfe7b3bfSKonstantin Belousov for (i = 0; i < 1000; i++) { 492dfe7b3bfSKonstantin Belousov if (sched_runnable()) 493dfe7b3bfSKonstantin Belousov return; 494dfe7b3bfSKonstantin Belousov cpu_spinwait(); 495dfe7b3bfSKonstantin Belousov } 496dfe7b3bfSKonstantin Belousov } 497dfe7b3bfSKonstantin Belousov 498dfe7b3bfSKonstantin Belousov /* 499dfe7b3bfSKonstantin Belousov * C1E renders the local APIC timer dead, so we disable it by 500dfe7b3bfSKonstantin Belousov * reading the Interrupt Pending Message register and clearing 501dfe7b3bfSKonstantin Belousov * both C1eOnCmpHalt (bit 28) and SmiOnCmpHalt (bit 27). 502dfe7b3bfSKonstantin Belousov * 503dfe7b3bfSKonstantin Belousov * Reference: 504dfe7b3bfSKonstantin Belousov * "BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh Processors" 505dfe7b3bfSKonstantin Belousov * #32559 revision 3.00+ 506dfe7b3bfSKonstantin Belousov */ 507dfe7b3bfSKonstantin Belousov #define MSR_AMDK8_IPM 0xc0010055 508dfe7b3bfSKonstantin Belousov #define AMDK8_SMIONCMPHALT (1ULL << 27) 509dfe7b3bfSKonstantin Belousov #define AMDK8_C1EONCMPHALT (1ULL << 28) 510dfe7b3bfSKonstantin Belousov #define AMDK8_CMPHALT (AMDK8_SMIONCMPHALT | AMDK8_C1EONCMPHALT) 511dfe7b3bfSKonstantin Belousov 512dfe7b3bfSKonstantin Belousov void 513dfe7b3bfSKonstantin Belousov cpu_probe_amdc1e(void) 514dfe7b3bfSKonstantin Belousov { 515dfe7b3bfSKonstantin Belousov 516dfe7b3bfSKonstantin Belousov /* 517dfe7b3bfSKonstantin Belousov * Detect the presence of C1E capability mostly on latest 518dfe7b3bfSKonstantin Belousov * dual-cores (or future) k8 family. 519dfe7b3bfSKonstantin Belousov */ 520dfe7b3bfSKonstantin Belousov if (cpu_vendor_id == CPU_VENDOR_AMD && 521dfe7b3bfSKonstantin Belousov (cpu_id & 0x00000f00) == 0x00000f00 && 522dfe7b3bfSKonstantin Belousov (cpu_id & 0x0fff0000) >= 0x00040000) { 523dfe7b3bfSKonstantin Belousov cpu_ident_amdc1e = 1; 524dfe7b3bfSKonstantin Belousov } 525dfe7b3bfSKonstantin Belousov } 526dfe7b3bfSKonstantin Belousov 527dfe7b3bfSKonstantin Belousov void (*cpu_idle_fn)(sbintime_t) = cpu_idle_acpi; 528dfe7b3bfSKonstantin Belousov 529dfe7b3bfSKonstantin Belousov void 530dfe7b3bfSKonstantin Belousov cpu_idle(int busy) 531dfe7b3bfSKonstantin Belousov { 532dfe7b3bfSKonstantin Belousov uint64_t msr; 533dfe7b3bfSKonstantin Belousov sbintime_t sbt = -1; 534dfe7b3bfSKonstantin Belousov 535dfe7b3bfSKonstantin Belousov CTR2(KTR_SPARE2, "cpu_idle(%d) at %d", 536dfe7b3bfSKonstantin Belousov busy, curcpu); 537ed95805eSJohn Baldwin #ifdef MP_WATCHDOG 538dfe7b3bfSKonstantin Belousov ap_watchdog(PCPU_GET(cpuid)); 539dfe7b3bfSKonstantin Belousov #endif 540ed95805eSJohn Baldwin 541dfe7b3bfSKonstantin Belousov /* If we are busy - try to use fast methods. */ 542dfe7b3bfSKonstantin Belousov if (busy) { 543dfe7b3bfSKonstantin Belousov if ((cpu_feature2 & CPUID2_MON) && idle_mwait) { 544dfe7b3bfSKonstantin Belousov cpu_idle_mwait(busy); 545dfe7b3bfSKonstantin Belousov goto out; 546dfe7b3bfSKonstantin Belousov } 547dfe7b3bfSKonstantin Belousov } 548dfe7b3bfSKonstantin Belousov 549dfe7b3bfSKonstantin Belousov /* If we have time - switch timers into idle mode. */ 550dfe7b3bfSKonstantin Belousov if (!busy) { 551dfe7b3bfSKonstantin Belousov critical_enter(); 552dfe7b3bfSKonstantin Belousov sbt = cpu_idleclock(); 553dfe7b3bfSKonstantin Belousov } 554dfe7b3bfSKonstantin Belousov 555dfe7b3bfSKonstantin Belousov /* Apply AMD APIC timer C1E workaround. */ 556dfe7b3bfSKonstantin Belousov if (cpu_ident_amdc1e && cpu_disable_c3_sleep) { 557dfe7b3bfSKonstantin Belousov msr = rdmsr(MSR_AMDK8_IPM); 558dfe7b3bfSKonstantin Belousov if (msr & AMDK8_CMPHALT) 559dfe7b3bfSKonstantin Belousov wrmsr(MSR_AMDK8_IPM, msr & ~AMDK8_CMPHALT); 560dfe7b3bfSKonstantin Belousov } 561dfe7b3bfSKonstantin Belousov 562dfe7b3bfSKonstantin Belousov /* Call main idle method. */ 563dfe7b3bfSKonstantin Belousov cpu_idle_fn(sbt); 564dfe7b3bfSKonstantin Belousov 565dfe7b3bfSKonstantin Belousov /* Switch timers back into active mode. */ 566dfe7b3bfSKonstantin Belousov if (!busy) { 567dfe7b3bfSKonstantin Belousov cpu_activeclock(); 568dfe7b3bfSKonstantin Belousov critical_exit(); 569dfe7b3bfSKonstantin Belousov } 570dfe7b3bfSKonstantin Belousov out: 571dfe7b3bfSKonstantin Belousov CTR2(KTR_SPARE2, "cpu_idle(%d) at %d done", 572dfe7b3bfSKonstantin Belousov busy, curcpu); 573dfe7b3bfSKonstantin Belousov } 574dfe7b3bfSKonstantin Belousov 575dfe7b3bfSKonstantin Belousov int 576dfe7b3bfSKonstantin Belousov cpu_idle_wakeup(int cpu) 577dfe7b3bfSKonstantin Belousov { 578dfe7b3bfSKonstantin Belousov struct pcpu *pcpu; 579dfe7b3bfSKonstantin Belousov int *state; 580dfe7b3bfSKonstantin Belousov 581dfe7b3bfSKonstantin Belousov pcpu = pcpu_find(cpu); 582dfe7b3bfSKonstantin Belousov state = (int *)pcpu->pc_monitorbuf; 583dfe7b3bfSKonstantin Belousov /* 584dfe7b3bfSKonstantin Belousov * This doesn't need to be atomic since missing the race will 585dfe7b3bfSKonstantin Belousov * simply result in unnecessary IPIs. 586dfe7b3bfSKonstantin Belousov */ 587dfe7b3bfSKonstantin Belousov if (*state == STATE_SLEEPING) 588dfe7b3bfSKonstantin Belousov return (0); 589dfe7b3bfSKonstantin Belousov if (*state == STATE_MWAIT) 590dfe7b3bfSKonstantin Belousov *state = STATE_RUNNING; 591dfe7b3bfSKonstantin Belousov return (1); 592dfe7b3bfSKonstantin Belousov } 593dfe7b3bfSKonstantin Belousov 594dfe7b3bfSKonstantin Belousov /* 595dfe7b3bfSKonstantin Belousov * Ordered by speed/power consumption. 596dfe7b3bfSKonstantin Belousov */ 597dfe7b3bfSKonstantin Belousov struct { 598dfe7b3bfSKonstantin Belousov void *id_fn; 599dfe7b3bfSKonstantin Belousov char *id_name; 600dfe7b3bfSKonstantin Belousov } idle_tbl[] = { 601dfe7b3bfSKonstantin Belousov { cpu_idle_spin, "spin" }, 602dfe7b3bfSKonstantin Belousov { cpu_idle_mwait, "mwait" }, 603dfe7b3bfSKonstantin Belousov { cpu_idle_hlt, "hlt" }, 604dfe7b3bfSKonstantin Belousov { cpu_idle_acpi, "acpi" }, 605dfe7b3bfSKonstantin Belousov { NULL, NULL } 606dfe7b3bfSKonstantin Belousov }; 607dfe7b3bfSKonstantin Belousov 608dfe7b3bfSKonstantin Belousov static int 609dfe7b3bfSKonstantin Belousov idle_sysctl_available(SYSCTL_HANDLER_ARGS) 610dfe7b3bfSKonstantin Belousov { 611dfe7b3bfSKonstantin Belousov char *avail, *p; 612dfe7b3bfSKonstantin Belousov int error; 613dfe7b3bfSKonstantin Belousov int i; 614dfe7b3bfSKonstantin Belousov 615dfe7b3bfSKonstantin Belousov avail = malloc(256, M_TEMP, M_WAITOK); 616dfe7b3bfSKonstantin Belousov p = avail; 617dfe7b3bfSKonstantin Belousov for (i = 0; idle_tbl[i].id_name != NULL; i++) { 618dfe7b3bfSKonstantin Belousov if (strstr(idle_tbl[i].id_name, "mwait") && 619dfe7b3bfSKonstantin Belousov (cpu_feature2 & CPUID2_MON) == 0) 620dfe7b3bfSKonstantin Belousov continue; 621dfe7b3bfSKonstantin Belousov if (strcmp(idle_tbl[i].id_name, "acpi") == 0 && 622dfe7b3bfSKonstantin Belousov cpu_idle_hook == NULL) 623dfe7b3bfSKonstantin Belousov continue; 624dfe7b3bfSKonstantin Belousov p += sprintf(p, "%s%s", p != avail ? ", " : "", 625dfe7b3bfSKonstantin Belousov idle_tbl[i].id_name); 626dfe7b3bfSKonstantin Belousov } 627dfe7b3bfSKonstantin Belousov error = sysctl_handle_string(oidp, avail, 0, req); 628dfe7b3bfSKonstantin Belousov free(avail, M_TEMP); 629dfe7b3bfSKonstantin Belousov return (error); 630dfe7b3bfSKonstantin Belousov } 631dfe7b3bfSKonstantin Belousov 632dfe7b3bfSKonstantin Belousov SYSCTL_PROC(_machdep, OID_AUTO, idle_available, CTLTYPE_STRING | CTLFLAG_RD, 633dfe7b3bfSKonstantin Belousov 0, 0, idle_sysctl_available, "A", "list of available idle functions"); 634dfe7b3bfSKonstantin Belousov 635dfe7b3bfSKonstantin Belousov static int 636dfe7b3bfSKonstantin Belousov idle_sysctl(SYSCTL_HANDLER_ARGS) 637dfe7b3bfSKonstantin Belousov { 638dfe7b3bfSKonstantin Belousov char buf[16]; 639dfe7b3bfSKonstantin Belousov int error; 640dfe7b3bfSKonstantin Belousov char *p; 641dfe7b3bfSKonstantin Belousov int i; 642dfe7b3bfSKonstantin Belousov 643dfe7b3bfSKonstantin Belousov p = "unknown"; 644dfe7b3bfSKonstantin Belousov for (i = 0; idle_tbl[i].id_name != NULL; i++) { 645dfe7b3bfSKonstantin Belousov if (idle_tbl[i].id_fn == cpu_idle_fn) { 646dfe7b3bfSKonstantin Belousov p = idle_tbl[i].id_name; 647dfe7b3bfSKonstantin Belousov break; 648dfe7b3bfSKonstantin Belousov } 649dfe7b3bfSKonstantin Belousov } 650dfe7b3bfSKonstantin Belousov strncpy(buf, p, sizeof(buf)); 651dfe7b3bfSKonstantin Belousov error = sysctl_handle_string(oidp, buf, sizeof(buf), req); 652dfe7b3bfSKonstantin Belousov if (error != 0 || req->newptr == NULL) 653dfe7b3bfSKonstantin Belousov return (error); 654dfe7b3bfSKonstantin Belousov for (i = 0; idle_tbl[i].id_name != NULL; i++) { 655dfe7b3bfSKonstantin Belousov if (strstr(idle_tbl[i].id_name, "mwait") && 656dfe7b3bfSKonstantin Belousov (cpu_feature2 & CPUID2_MON) == 0) 657dfe7b3bfSKonstantin Belousov continue; 658dfe7b3bfSKonstantin Belousov if (strcmp(idle_tbl[i].id_name, "acpi") == 0 && 659dfe7b3bfSKonstantin Belousov cpu_idle_hook == NULL) 660dfe7b3bfSKonstantin Belousov continue; 661dfe7b3bfSKonstantin Belousov if (strcmp(idle_tbl[i].id_name, buf)) 662dfe7b3bfSKonstantin Belousov continue; 663dfe7b3bfSKonstantin Belousov cpu_idle_fn = idle_tbl[i].id_fn; 664dfe7b3bfSKonstantin Belousov return (0); 665dfe7b3bfSKonstantin Belousov } 666dfe7b3bfSKonstantin Belousov return (EINVAL); 667dfe7b3bfSKonstantin Belousov } 668dfe7b3bfSKonstantin Belousov 669dfe7b3bfSKonstantin Belousov SYSCTL_PROC(_machdep, OID_AUTO, idle, CTLTYPE_STRING | CTLFLAG_RW, 0, 0, 670dfe7b3bfSKonstantin Belousov idle_sysctl, "A", "currently selected idle function"); 671835c2787SKonstantin Belousov 672295f4b6cSKonstantin Belousov static int panic_on_nmi = 1; 673295f4b6cSKonstantin Belousov SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RWTUN, 674295f4b6cSKonstantin Belousov &panic_on_nmi, 0, 675295f4b6cSKonstantin Belousov "Panic on NMI"); 676835c2787SKonstantin Belousov int nmi_is_broadcast = 1; 677835c2787SKonstantin Belousov SYSCTL_INT(_machdep, OID_AUTO, nmi_is_broadcast, CTLFLAG_RWTUN, 678835c2787SKonstantin Belousov &nmi_is_broadcast, 0, 679835c2787SKonstantin Belousov "Chipset NMI is broadcast"); 680835c2787SKonstantin Belousov #ifdef KDB 681835c2787SKonstantin Belousov int kdb_on_nmi = 1; 682835c2787SKonstantin Belousov SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RWTUN, 683835c2787SKonstantin Belousov &kdb_on_nmi, 0, 684835c2787SKonstantin Belousov "Go to KDB on NMI"); 685835c2787SKonstantin Belousov #endif 686835c2787SKonstantin Belousov 687835c2787SKonstantin Belousov #ifdef DEV_ISA 688295f4b6cSKonstantin Belousov void 689295f4b6cSKonstantin Belousov nmi_call_kdb(u_int cpu, u_int type, struct trapframe *frame) 690835c2787SKonstantin Belousov { 691835c2787SKonstantin Belousov 692835c2787SKonstantin Belousov /* machine/parity/power fail/"kitchen sink" faults */ 693835c2787SKonstantin Belousov if (isa_nmi(frame->tf_err) == 0) { 694835c2787SKonstantin Belousov #ifdef KDB 695835c2787SKonstantin Belousov /* 696835c2787SKonstantin Belousov * NMI can be hooked up to a pushbutton for debugging. 697835c2787SKonstantin Belousov */ 698835c2787SKonstantin Belousov if (kdb_on_nmi) { 699835c2787SKonstantin Belousov printf("NMI/cpu%d ... going to debugger\n", cpu); 700835c2787SKonstantin Belousov kdb_trap(type, 0, frame); 701835c2787SKonstantin Belousov } 702835c2787SKonstantin Belousov #endif /* KDB */ 703295f4b6cSKonstantin Belousov } else if (panic_on_nmi) { 704835c2787SKonstantin Belousov panic("NMI indicates hardware failure"); 705295f4b6cSKonstantin Belousov } 706835c2787SKonstantin Belousov } 707835c2787SKonstantin Belousov #endif 708835c2787SKonstantin Belousov 709295f4b6cSKonstantin Belousov void 710295f4b6cSKonstantin Belousov nmi_handle_intr(u_int type, struct trapframe *frame) 711835c2787SKonstantin Belousov { 712835c2787SKonstantin Belousov 713835c2787SKonstantin Belousov #ifdef DEV_ISA 714835c2787SKonstantin Belousov #ifdef SMP 715295f4b6cSKonstantin Belousov if (nmi_is_broadcast) { 716295f4b6cSKonstantin Belousov nmi_call_kdb_smp(type, frame); 717295f4b6cSKonstantin Belousov return; 718295f4b6cSKonstantin Belousov } 719835c2787SKonstantin Belousov #endif 7201d6dfd12SKonstantin Belousov nmi_call_kdb(PCPU_GET(cpuid), type, frame); 721835c2787SKonstantin Belousov #endif 722835c2787SKonstantin Belousov } 723319117fdSKonstantin Belousov 724319117fdSKonstantin Belousov int hw_ibrs_active; 725319117fdSKonstantin Belousov int hw_ibrs_disable = 1; 726319117fdSKonstantin Belousov 727319117fdSKonstantin Belousov SYSCTL_INT(_hw, OID_AUTO, ibrs_active, CTLFLAG_RD, &hw_ibrs_active, 0, 728b31b965eSKonstantin Belousov "Indirect Branch Restricted Speculation active"); 729319117fdSKonstantin Belousov 730319117fdSKonstantin Belousov void 731319117fdSKonstantin Belousov hw_ibrs_recalculate(void) 732319117fdSKonstantin Belousov { 733319117fdSKonstantin Belousov uint64_t v; 734319117fdSKonstantin Belousov 735319117fdSKonstantin Belousov if ((cpu_ia32_arch_caps & IA32_ARCH_CAP_IBRS_ALL) != 0) { 736319117fdSKonstantin Belousov if (hw_ibrs_disable) { 737319117fdSKonstantin Belousov v= rdmsr(MSR_IA32_SPEC_CTRL); 738c688c905SKonstantin Belousov v &= ~(uint64_t)IA32_SPEC_CTRL_IBRS; 739319117fdSKonstantin Belousov wrmsr(MSR_IA32_SPEC_CTRL, v); 740319117fdSKonstantin Belousov } else { 741319117fdSKonstantin Belousov v= rdmsr(MSR_IA32_SPEC_CTRL); 742319117fdSKonstantin Belousov v |= IA32_SPEC_CTRL_IBRS; 743319117fdSKonstantin Belousov wrmsr(MSR_IA32_SPEC_CTRL, v); 744319117fdSKonstantin Belousov } 745319117fdSKonstantin Belousov return; 746319117fdSKonstantin Belousov } 747319117fdSKonstantin Belousov hw_ibrs_active = (cpu_stdext_feature3 & CPUID_STDEXT3_IBPB) != 0 && 748319117fdSKonstantin Belousov !hw_ibrs_disable; 749319117fdSKonstantin Belousov } 750319117fdSKonstantin Belousov 751319117fdSKonstantin Belousov static int 752319117fdSKonstantin Belousov hw_ibrs_disable_handler(SYSCTL_HANDLER_ARGS) 753319117fdSKonstantin Belousov { 754319117fdSKonstantin Belousov int error, val; 755319117fdSKonstantin Belousov 756319117fdSKonstantin Belousov val = hw_ibrs_disable; 757319117fdSKonstantin Belousov error = sysctl_handle_int(oidp, &val, 0, req); 758319117fdSKonstantin Belousov if (error != 0 || req->newptr == NULL) 759319117fdSKonstantin Belousov return (error); 760319117fdSKonstantin Belousov hw_ibrs_disable = val != 0; 761319117fdSKonstantin Belousov hw_ibrs_recalculate(); 762319117fdSKonstantin Belousov return (0); 763319117fdSKonstantin Belousov } 764319117fdSKonstantin Belousov SYSCTL_PROC(_hw, OID_AUTO, ibrs_disable, CTLTYPE_INT | CTLFLAG_RWTUN | 765319117fdSKonstantin Belousov CTLFLAG_NOFETCH | CTLFLAG_MPSAFE, NULL, 0, hw_ibrs_disable_handler, "I", 766b31b965eSKonstantin Belousov "Disable Indirect Branch Restricted Speculation"); 7678fbcc334SKonstantin Belousov 7688fbcc334SKonstantin Belousov /* 7698fbcc334SKonstantin Belousov * Enable and restore kernel text write permissions. 7708fbcc334SKonstantin Belousov * Callers must ensure that disable_wp()/restore_wp() are executed 7718fbcc334SKonstantin Belousov * without rescheduling on the same core. 7728fbcc334SKonstantin Belousov */ 7738fbcc334SKonstantin Belousov bool 7748fbcc334SKonstantin Belousov disable_wp(void) 7758fbcc334SKonstantin Belousov { 7768fbcc334SKonstantin Belousov u_int cr0; 7778fbcc334SKonstantin Belousov 7788fbcc334SKonstantin Belousov cr0 = rcr0(); 7798fbcc334SKonstantin Belousov if ((cr0 & CR0_WP) == 0) 7808fbcc334SKonstantin Belousov return (false); 7818fbcc334SKonstantin Belousov load_cr0(cr0 & ~CR0_WP); 7828fbcc334SKonstantin Belousov return (true); 7838fbcc334SKonstantin Belousov } 7848fbcc334SKonstantin Belousov 7858fbcc334SKonstantin Belousov void 7868fbcc334SKonstantin Belousov restore_wp(bool old_wp) 7878fbcc334SKonstantin Belousov { 7888fbcc334SKonstantin Belousov 7898fbcc334SKonstantin Belousov if (old_wp) 7908fbcc334SKonstantin Belousov load_cr0(rcr0() | CR0_WP); 7918fbcc334SKonstantin Belousov } 7928fbcc334SKonstantin Belousov 793