diff --git a/sys/ia64/ia64/clock.c b/sys/ia64/ia64/clock.c index 22fcf9bf452c..6f2e2d86683e 100644 --- a/sys/ia64/ia64/clock.c +++ b/sys/ia64/ia64/clock.c @@ -1,108 +1,108 @@ /*- * Copyright (c) 2005 Marcel Moolenaar * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include uint64_t ia64_clock_reload; #ifndef SMP static timecounter_get_t ia64_get_timecount; static struct timecounter ia64_timecounter = { ia64_get_timecount, /* get_timecount */ 0, /* no poll_pps */ ~0u, /* counter_mask */ 0, /* frequency */ "ITC" /* name */ }; static unsigned ia64_get_timecount(struct timecounter* tc) { return ia64_get_itc(); } #endif void pcpu_initclock(void) { - PCPU_SET(clockadj, 0); - PCPU_SET(clock, ia64_get_itc()); - ia64_set_itm(PCPU_GET(clock) + ia64_clock_reload); + PCPU_SET(md.clockadj, 0); + PCPU_SET(md.clock, ia64_get_itc()); + ia64_set_itm(PCPU_GET(md.clock) + ia64_clock_reload); ia64_set_itv(CLOCK_VECTOR); /* highest priority class */ ia64_srlz_d(); } /* * Start the real-time and statistics clocks. We use cr.itc and cr.itm * to implement a 1000hz clock. */ void cpu_initclocks() { if (itc_frequency == 0) panic("Unknown clock frequency"); stathz = hz; ia64_clock_reload = (itc_frequency + hz/2) / hz; #ifndef SMP ia64_timecounter.tc_frequency = itc_frequency; tc_init(&ia64_timecounter); #endif pcpu_initclock(); } void cpu_startprofclock(void) { /* nothing to do */ } void cpu_stopprofclock(void) { /* nothing to do */ } diff --git a/sys/ia64/ia64/genassym.c b/sys/ia64/ia64/genassym.c index 4a192fd79e39..9f7625279d9c 100644 --- a/sys/ia64/ia64/genassym.c +++ b/sys/ia64/ia64/genassym.c @@ -1,120 +1,120 @@ /*- * Copyright (c) 1982, 1990 The Regents of the University of California. * All rights reserved. * * This code is derived from software contributed to Berkeley by * William Jolitz. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include "opt_compat.h" #include "opt_kstack_pages.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef COMPAT_IA32 ASSYM(COMPAT_IA32, COMPAT_IA32); #endif ASSYM(DT_NULL, DT_NULL); ASSYM(DT_RELA, DT_RELA); ASSYM(DT_RELAENT, DT_RELAENT); ASSYM(DT_RELASZ, DT_RELASZ); ASSYM(DT_SYMTAB, DT_SYMTAB); ASSYM(DT_SYMENT, DT_SYMENT); ASSYM(EFAULT, EFAULT); ASSYM(ENAMETOOLONG, ENAMETOOLONG); ASSYM(ERESTART, ERESTART); ASSYM(FRAME_SYSCALL, FRAME_SYSCALL); ASSYM(IA64_ID_PAGE_SHIFT, IA64_ID_PAGE_SHIFT); ASSYM(KSTACK_PAGES, KSTACK_PAGES); ASSYM(MC_PRESERVED, offsetof(mcontext_t, mc_preserved)); ASSYM(MC_PRESERVED_FP, offsetof(mcontext_t, mc_preserved_fp)); ASSYM(MC_SPECIAL, offsetof(mcontext_t, mc_special)); ASSYM(MC_SPECIAL_BSPSTORE, offsetof(mcontext_t, mc_special.bspstore)); ASSYM(MC_SPECIAL_RNAT, offsetof(mcontext_t, mc_special.rnat)); ASSYM(PAGE_SHIFT, PAGE_SHIFT); ASSYM(PAGE_SIZE, PAGE_SIZE); -ASSYM(PC_CURRENT_PMAP, offsetof(struct pcpu, pc_current_pmap)); +ASSYM(PC_CURRENT_PMAP, offsetof(struct pcpu, pc_md.current_pmap)); ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread)); ASSYM(PC_IDLETHREAD, offsetof(struct pcpu, pc_idlethread)); ASSYM(PCB_CURRENT_PMAP, offsetof(struct pcb, pcb_current_pmap)); ASSYM(PCB_ONFAULT, offsetof(struct pcb, pcb_onfault)); ASSYM(PCB_SPECIAL_RP, offsetof(struct pcb, pcb_special.rp)); ASSYM(R_IA_64_DIR64LSB, R_IA_64_DIR64LSB); ASSYM(R_IA_64_FPTR64LSB, R_IA_64_FPTR64LSB); ASSYM(R_IA_64_NONE, R_IA_64_NONE); ASSYM(R_IA_64_REL64LSB, R_IA_64_REL64LSB); ASSYM(SIZEOF_PCB, sizeof(struct pcb)); ASSYM(SIZEOF_SPECIAL, sizeof(struct _special)); ASSYM(SIZEOF_TRAPFRAME, sizeof(struct trapframe)); ASSYM(TD_FLAGS, offsetof(struct thread, td_flags)); ASSYM(TD_KSTACK, offsetof(struct thread, td_kstack)); ASSYM(TD_PCB, offsetof(struct thread, td_pcb)); ASSYM(TDF_ASTPENDING, TDF_ASTPENDING); ASSYM(TDF_NEEDRESCHED, TDF_NEEDRESCHED); ASSYM(UC_MCONTEXT, offsetof(ucontext_t, uc_mcontext)); ASSYM(VM_MAX_ADDRESS, VM_MAX_ADDRESS); diff --git a/sys/ia64/ia64/interrupt.c b/sys/ia64/ia64/interrupt.c index 7bd6694a343f..8a8cbb81a953 100644 --- a/sys/ia64/ia64/interrupt.c +++ b/sys/ia64/ia64/interrupt.c @@ -1,427 +1,427 @@ /* $FreeBSD$ */ /* $NetBSD: interrupt.c,v 1.23 1998/02/24 07:38:01 thorpej Exp $ */ /*- * Copyright (c) 1994, 1995, 1996 Carnegie-Mellon University. * All rights reserved. * * Authors: Keith Bostic, Chris G. Demetriou * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /*- * Additional Copyright (c) 1997 by Matthew Jacob for NASA/Ames Research Center. * Redistribute and modify at will, leaving only this additional copyright * notice. */ #include "opt_ddb.h" #include /* RCS ID & Copyright macro defns */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef EVCNT_COUNTERS struct evcnt clock_intr_evcnt; /* event counter for clock intrs. */ #else #include #include #endif #ifdef DDB #include #endif static void ia64_dispatch_intr(void *, u_int); static void dummy_perf(unsigned long vector, struct trapframe *tf) { printf("performance interrupt!\n"); } void (*perf_irq)(unsigned long, struct trapframe *) = dummy_perf; SYSCTL_NODE(_debug, OID_AUTO, clock, CTLFLAG_RW, 0, "clock statistics"); static int adjust_edges = 0; SYSCTL_INT(_debug_clock, OID_AUTO, adjust_edges, CTLFLAG_RD, &adjust_edges, 0, "Number of times ITC got more than 12.5% behind"); static int adjust_excess = 0; SYSCTL_INT(_debug_clock, OID_AUTO, adjust_excess, CTLFLAG_RD, &adjust_excess, 0, "Total number of ignored ITC interrupts"); static int adjust_lost = 0; SYSCTL_INT(_debug_clock, OID_AUTO, adjust_lost, CTLFLAG_RD, &adjust_lost, 0, "Total number of lost ITC interrupts"); static int adjust_ticks = 0; SYSCTL_INT(_debug_clock, OID_AUTO, adjust_ticks, CTLFLAG_RD, &adjust_ticks, 0, "Total number of ITC interrupts with adjustment"); void interrupt(struct trapframe *tf) { struct thread *td; volatile struct ia64_interrupt_block *ib = IA64_INTERRUPT_BLOCK; uint64_t adj, clk, itc; int64_t delta; u_int vector; int count; uint8_t inta; ia64_set_fpsr(IA64_FPSR_DEFAULT); td = curthread; PCPU_INC(cnt.v_intr); vector = tf->tf_special.ifa; next: /* * Handle ExtINT interrupts by generating an INTA cycle to * read the vector. */ if (vector == 0) { - PCPU_INC(stats.pcs_nextints); + PCPU_INC(md.stats.pcs_nextints); inta = ib->ib_inta; if (inta == 15) { - PCPU_INC(stats.pcs_nstrays); + PCPU_INC(md.stats.pcs_nstrays); __asm __volatile("mov cr.eoi = r0;; srlz.d"); goto stray; } vector = (int)inta; } else if (vector == 15) { - PCPU_INC(stats.pcs_nstrays); + PCPU_INC(md.stats.pcs_nstrays); goto stray; } if (vector == CLOCK_VECTOR) {/* clock interrupt */ /* CTR0(KTR_INTR, "clock interrupt"); */ itc = ia64_get_itc(); - PCPU_INC(stats.pcs_nclks); + PCPU_INC(md.stats.pcs_nclks); #ifdef EVCNT_COUNTERS clock_intr_evcnt.ev_count++; #else intrcnt[INTRCNT_CLOCK]++; #endif critical_enter(); - adj = PCPU_GET(clockadj); - clk = PCPU_GET(clock); + adj = PCPU_GET(md.clockadj); + clk = PCPU_GET(md.clock); delta = itc - clk; count = 0; while (delta >= ia64_clock_reload) { /* Only the BSP runs the real clock */ if (PCPU_GET(cpuid) == 0) hardclock(TRAPF_USERMODE(tf), TRAPF_PC(tf)); else hardclock_cpu(TRAPF_USERMODE(tf)); if (profprocs != 0) profclock(TRAPF_USERMODE(tf), TRAPF_PC(tf)); statclock(TRAPF_USERMODE(tf)); delta -= ia64_clock_reload; clk += ia64_clock_reload; if (adj != 0) adjust_ticks++; count++; } ia64_set_itm(ia64_get_itc() + ia64_clock_reload - adj); if (count > 0) { adjust_lost += count - 1; if (delta > (ia64_clock_reload >> 3)) { if (adj == 0) adjust_edges++; adj = ia64_clock_reload >> 4; } else adj = 0; } else { adj = 0; adjust_excess++; } - PCPU_SET(clock, clk); - PCPU_SET(clockadj, adj); + PCPU_SET(md.clock, clk); + PCPU_SET(md.clockadj, adj); critical_exit(); ia64_srlz_d(); #ifdef SMP } else if (vector == ipi_vector[IPI_AST]) { - PCPU_INC(stats.pcs_nasts); + PCPU_INC(md.stats.pcs_nasts); CTR1(KTR_SMP, "IPI_AST, cpuid=%d", PCPU_GET(cpuid)); } else if (vector == ipi_vector[IPI_HIGH_FP]) { - PCPU_INC(stats.pcs_nhighfps); + PCPU_INC(md.stats.pcs_nhighfps); ia64_highfp_save_ipi(); } else if (vector == ipi_vector[IPI_RENDEZVOUS]) { - PCPU_INC(stats.pcs_nrdvs); + PCPU_INC(md.stats.pcs_nrdvs); CTR1(KTR_SMP, "IPI_RENDEZVOUS, cpuid=%d", PCPU_GET(cpuid)); enable_intr(); smp_rendezvous_action(); disable_intr(); } else if (vector == ipi_vector[IPI_STOP]) { - PCPU_INC(stats.pcs_nstops); + PCPU_INC(md.stats.pcs_nstops); cpumask_t mybit = PCPU_GET(cpumask); /* Make sure IPI_STOP_HARD is mapped to IPI_STOP. */ KASSERT(IPI_STOP == IPI_STOP_HARD, ("%s: IPI_STOP_HARD not handled.", __func__)); - savectx(PCPU_PTR(pcb)); + savectx(PCPU_PTR(md.pcb)); atomic_set_int(&stopped_cpus, mybit); while ((started_cpus & mybit) == 0) cpu_spinwait(); atomic_clear_int(&started_cpus, mybit); atomic_clear_int(&stopped_cpus, mybit); } else if (vector == ipi_vector[IPI_PREEMPT]) { - PCPU_INC(stats.pcs_npreempts); + PCPU_INC(md.stats.pcs_npreempts); CTR1(KTR_SMP, "IPI_PREEMPT, cpuid=%d", PCPU_GET(cpuid)); __asm __volatile("mov cr.eoi = r0;; srlz.d"); enable_intr(); sched_preempt(curthread); disable_intr(); goto stray; #endif } else { - PCPU_INC(stats.pcs_nhwints); + PCPU_INC(md.stats.pcs_nhwints); atomic_add_int(&td->td_intr_nesting_level, 1); ia64_dispatch_intr(tf, vector); atomic_subtract_int(&td->td_intr_nesting_level, 1); } __asm __volatile("mov cr.eoi = r0;; srlz.d"); vector = ia64_get_ivr(); if (vector != 15) goto next; stray: if (TRAPF_USERMODE(tf)) { enable_intr(); userret(td, tf); mtx_assert(&Giant, MA_NOTOWNED); do_ast(tf); } } /* * Hardware irqs have vectors starting at this offset. */ #define IA64_HARDWARE_IRQ_BASE 0x20 struct ia64_intr { struct intr_event *event; /* interrupt event */ volatile long *cntp; /* interrupt counter */ struct sapic *sapic; u_int irq; }; static struct ia64_intr *ia64_intrs[256]; static void ia64_intr_eoi(void *arg) { u_int vector = (uintptr_t)arg; struct ia64_intr *i; i = ia64_intrs[vector]; if (i != NULL) sapic_eoi(i->sapic, vector); } static void ia64_intr_mask(void *arg) { u_int vector = (uintptr_t)arg; struct ia64_intr *i; i = ia64_intrs[vector]; if (i != NULL) { sapic_mask(i->sapic, i->irq); sapic_eoi(i->sapic, vector); } } static void ia64_intr_unmask(void *arg) { u_int vector = (uintptr_t)arg; struct ia64_intr *i; i = ia64_intrs[vector]; if (i != NULL) sapic_unmask(i->sapic, i->irq); } int ia64_setup_intr(const char *name, int irq, driver_filter_t filter, driver_intr_t handler, void *arg, enum intr_type flags, void **cookiep) { struct ia64_intr *i; struct sapic *sa; char *intrname; u_int vector; int error; /* Get the I/O SAPIC that corresponds to the IRQ. */ sa = sapic_lookup(irq); if (sa == NULL) return (EINVAL); /* * XXX - There's a priority implied by the choice of vector. * We should therefore relate the vector to the interrupt type. */ vector = irq + IA64_HARDWARE_IRQ_BASE; i = ia64_intrs[vector]; if (i == NULL) { i = malloc(sizeof(struct ia64_intr), M_DEVBUF, M_NOWAIT); if (i == NULL) return (ENOMEM); error = intr_event_create(&i->event, (void *)(uintptr_t)vector, 0, irq, ia64_intr_mask, ia64_intr_unmask, ia64_intr_eoi, NULL, "irq%u:", irq); if (error) { free(i, M_DEVBUF); return (error); } if (!atomic_cmpset_ptr(&ia64_intrs[vector], NULL, i)) { intr_event_destroy(i->event); free(i, M_DEVBUF); i = ia64_intrs[vector]; } else { i->sapic = sa; i->irq = irq; i->cntp = intrcnt + irq + INTRCNT_ISA_IRQ; if (name != NULL && *name != '\0') { /* XXX needs abstraction. Too error prone. */ intrname = intrnames + (irq + INTRCNT_ISA_IRQ) * INTRNAME_LEN; memset(intrname, ' ', INTRNAME_LEN - 1); bcopy(name, intrname, strlen(name)); } sapic_enable(i->sapic, irq, vector); } } error = intr_event_add_handler(i->event, name, filter, handler, arg, intr_priority(flags), flags, cookiep); return (error); } int ia64_teardown_intr(void *cookie) { return (intr_event_remove_handler(cookie)); } static void ia64_dispatch_intr(void *frame, u_int vector) { struct ia64_intr *i; struct intr_event *ie; /* our interrupt event */ /* * Find the interrupt thread for this vector. */ i = ia64_intrs[vector]; KASSERT(i != NULL, ("%s: unassigned vector", __func__)); (*i->cntp)++; ie = i->event; KASSERT(ie != NULL, ("%s: interrupt without event", __func__)); if (intr_event_handle(ie, frame) != 0) { /* * XXX: The pre-INTR_FILTER code didn't mask stray * interrupts. */ ia64_intr_mask((void *)(uintptr_t)vector); log(LOG_ERR, "stray irq%u\n", i->irq); } } #ifdef DDB static void db_print_vector(u_int vector, int always) { struct ia64_intr *i; i = ia64_intrs[vector]; if (i != NULL) { db_printf("vector %u (%p): ", vector, i); sapic_print(i->sapic, i->irq); } else if (always) db_printf("vector %u: unassigned\n", vector); } DB_SHOW_COMMAND(vector, db_show_vector) { u_int vector; if (have_addr) { vector = ((addr >> 4) % 16) * 10 + (addr % 16); if (vector >= 256) db_printf("error: vector %u not in range [0..255]\n", vector); else db_print_vector(vector, 1); } else { for (vector = 0; vector < 256; vector++) db_print_vector(vector, 0); } } #endif diff --git a/sys/ia64/ia64/machdep.c b/sys/ia64/ia64/machdep.c index 15ee5fdd2d41..6d34e4d26d80 100644 --- a/sys/ia64/ia64/machdep.c +++ b/sys/ia64/ia64/machdep.c @@ -1,1540 +1,1543 @@ /*- * Copyright (c) 2003,2004 Marcel Moolenaar * Copyright (c) 2000,2001 Doug Rabson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_ddb.h" #include "opt_kstack_pages.h" #include "opt_msgbuf.h" #include "opt_sched.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef SMP #include #endif #include #include #include +SYSCTL_NODE(_machdep, OID_AUTO, cpu, CTLFLAG_RD, 0, ""); + u_int64_t processor_frequency; u_int64_t bus_frequency; u_int64_t itc_frequency; int cold = 1; u_int64_t pa_bootinfo; struct bootinfo bootinfo; struct pcpu pcpu0; extern u_int64_t kernel_text[], _end[]; extern u_int64_t ia64_gateway_page[]; extern u_int64_t break_sigtramp[]; extern u_int64_t epc_sigtramp[]; struct fpswa_iface *fpswa_iface; u_int64_t ia64_pal_base; u_int64_t ia64_port_base; static int ia64_sync_icache_needed; char machine[] = MACHINE; SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, ""); static char cpu_model[64]; SYSCTL_STRING(_hw, HW_MODEL, model, CTLFLAG_RD, cpu_model, 0, "The CPU model name"); static char cpu_family[64]; SYSCTL_STRING(_hw, OID_AUTO, family, CTLFLAG_RD, cpu_family, 0, "The CPU family name"); #ifdef DDB extern vm_offset_t ksym_start, ksym_end; #endif struct msgbuf *msgbufp = NULL; /* Other subsystems (e.g., ACPI) can hook this later. */ void (*cpu_idle_hook)(void) = NULL; long Maxmem = 0; long realmem = 0; #define PHYSMAP_SIZE (2 * VM_PHYSSEG_MAX) vm_paddr_t phys_avail[PHYSMAP_SIZE + 2]; /* must be 2 less so 0 0 can signal end of chunks */ #define PHYS_AVAIL_ARRAY_END ((sizeof(phys_avail) / sizeof(vm_offset_t)) - 2) struct kva_md_info kmi; #define Mhz 1000000L #define Ghz (1000L*Mhz) static void identifycpu(void) { char vendor[17]; char *family_name, *model_name; u_int64_t features, tmp; int number, revision, model, family, archrev; /* * Assumes little-endian. */ *(u_int64_t *) &vendor[0] = ia64_get_cpuid(0); *(u_int64_t *) &vendor[8] = ia64_get_cpuid(1); vendor[16] = '\0'; tmp = ia64_get_cpuid(3); number = (tmp >> 0) & 0xff; revision = (tmp >> 8) & 0xff; model = (tmp >> 16) & 0xff; family = (tmp >> 24) & 0xff; archrev = (tmp >> 32) & 0xff; family_name = model_name = "unknown"; switch (family) { case 0x07: family_name = "Itanium"; model_name = "Merced"; break; case 0x1f: family_name = "Itanium 2"; switch (model) { case 0x00: model_name = "McKinley"; break; case 0x01: /* * Deerfield is a low-voltage variant based on the * Madison core. We need circumstantial evidence * (i.e. the clock frequency) to identify those. * Allow for roughly 1% error margin. */ tmp = processor_frequency >> 7; if ((processor_frequency - tmp) < 1*Ghz && (processor_frequency + tmp) >= 1*Ghz) model_name = "Deerfield"; else model_name = "Madison"; break; case 0x02: model_name = "Madison II"; break; } break; case 0x20: ia64_sync_icache_needed = 1; family_name = "Itanium 2"; switch (model) { case 0x00: model_name = "Montecito"; break; } break; } snprintf(cpu_family, sizeof(cpu_family), "%s", family_name); snprintf(cpu_model, sizeof(cpu_model), "%s", model_name); features = ia64_get_cpuid(4); printf("CPU: %s (", model_name); if (processor_frequency) { printf("%ld.%02ld-Mhz ", (processor_frequency + 4999) / Mhz, ((processor_frequency + 4999) / (Mhz/100)) % 100); } printf("%s)\n", family_name); printf(" Origin = \"%s\" Revision = %d\n", vendor, revision); printf(" Features = 0x%b\n", (u_int32_t) features, "\020" "\001LB" /* long branch (brl) instruction. */ "\002SD" /* Spontaneous deferral. */ "\003AO" /* 16-byte atomic operations (ld, st, cmpxchg). */ ); } static void cpu_startup(void *dummy) { char nodename[16]; struct pcpu *pc; struct pcpu_stats *pcs; /* * Good {morning,afternoon,evening,night}. */ identifycpu(); #ifdef PERFMON perfmon_init(); #endif printf("real memory = %ld (%ld MB)\n", ia64_ptob(Maxmem), ia64_ptob(Maxmem) / 1048576); realmem = Maxmem; /* * Display any holes after the first chunk of extended memory. */ if (bootverbose) { int indx; printf("Physical memory chunk(s):\n"); for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) { long size1 = phys_avail[indx + 1] - phys_avail[indx]; printf("0x%08lx - 0x%08lx, %ld bytes (%ld pages)\n", phys_avail[indx], phys_avail[indx + 1] - 1, size1, size1 >> PAGE_SHIFT); } } vm_ksubmap_init(&kmi); printf("avail memory = %ld (%ld MB)\n", ptoa(cnt.v_free_count), ptoa(cnt.v_free_count) / 1048576); if (fpswa_iface == NULL) printf("Warning: no FPSWA package supplied\n"); else printf("FPSWA Revision = 0x%lx, Entry = %p\n", (long)fpswa_iface->if_rev, (void *)fpswa_iface->if_fpswa); /* * Set up buffers, so they can be used to read disk labels. */ bufinit(); vm_pager_bufferinit(); /* * Traverse the MADT to discover IOSAPIC and Local SAPIC * information. */ ia64_probe_sapics(); ia64_mca_init(); /* * Create sysctl tree for per-CPU information. */ SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { - pcs = &pc->pc_stats; - snprintf(nodename, sizeof(nodename), "cpu%u", pc->pc_cpuid); - sysctl_ctx_init(&pcs->pcs_sysctl_ctx); - pcs->pcs_sysctl_tree = SYSCTL_ADD_NODE(&pcs->pcs_sysctl_ctx, - SYSCTL_STATIC_CHILDREN(_machdep), OID_AUTO, nodename, + snprintf(nodename, sizeof(nodename), "%u", pc->pc_cpuid); + sysctl_ctx_init(&pc->pc_md.sysctl_ctx); + pc->pc_md.sysctl_tree = SYSCTL_ADD_NODE(&pc->pc_md.sysctl_ctx, + SYSCTL_STATIC_CHILDREN(_machdep_cpu), OID_AUTO, nodename, CTLFLAG_RD, NULL, ""); - if (pcs->pcs_sysctl_tree == NULL) + if (pc->pc_md.sysctl_tree == NULL) continue; - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + pcs = &pc->pc_md.stats; + + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nasts", CTLFLAG_RD, &pcs->pcs_nasts, "Number of IPI_AST interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nclks", CTLFLAG_RD, &pcs->pcs_nclks, "Number of clock interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nextints", CTLFLAG_RD, &pcs->pcs_nextints, "Number of ExtINT interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nhighfps", CTLFLAG_RD, &pcs->pcs_nhighfps, "Number of IPI_HIGH_FP interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nhwints", CTLFLAG_RD, &pcs->pcs_nhwints, "Number of hardware (device) interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "npreempts", CTLFLAG_RD, &pcs->pcs_npreempts, "Number of IPI_PREEMPT interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nrdvs", CTLFLAG_RD, &pcs->pcs_nrdvs, "Number of IPI_RENDEZVOUS interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nstops", CTLFLAG_RD, &pcs->pcs_nstops, "Number of IPI_STOP interrupts"); - SYSCTL_ADD_ULONG(&pcs->pcs_sysctl_ctx, - SYSCTL_CHILDREN(pcs->pcs_sysctl_tree), OID_AUTO, + SYSCTL_ADD_ULONG(&pc->pc_md.sysctl_ctx, + SYSCTL_CHILDREN(pc->pc_md.sysctl_tree), OID_AUTO, "nstrays", CTLFLAG_RD, &pcs->pcs_nstrays, "Number of stray vectors"); } } SYSINIT(cpu_startup, SI_SUB_CPU, SI_ORDER_FIRST, cpu_startup, NULL); void cpu_boot(int howto) { efi_reset_system(); } void cpu_flush_dcache(void *ptr, size_t len) { vm_offset_t lim, va; va = (uintptr_t)ptr & ~31; lim = (uintptr_t)ptr + len; while (va < lim) { ia64_fc(va); va += 32; } ia64_srlz_d(); } /* Get current clock frequency for the given cpu id. */ int cpu_est_clockrate(int cpu_id, uint64_t *rate) { if (pcpu_find(cpu_id) == NULL || rate == NULL) return (EINVAL); *rate = processor_frequency; return (0); } void cpu_halt() { efi_reset_system(); } void cpu_idle(int busy) { struct ia64_pal_result res; if (cpu_idle_hook != NULL) (*cpu_idle_hook)(); else res = ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0); } int cpu_idle_wakeup(int cpu) { return (0); } void cpu_reset() { cpu_boot(0); } void cpu_switch(struct thread *old, struct thread *new, struct mtx *mtx) { struct pcb *oldpcb, *newpcb; oldpcb = old->td_pcb; #ifdef COMPAT_IA32 ia32_savectx(oldpcb); #endif if (PCPU_GET(fpcurthread) == old) old->td_frame->tf_special.psr |= IA64_PSR_DFH; if (!savectx(oldpcb)) { old->td_lock = mtx; #if defined(SCHED_ULE) && defined(SMP) /* td_lock is volatile */ while (new->td_lock == &blocked_lock) ; #endif newpcb = new->td_pcb; oldpcb->pcb_current_pmap = pmap_switch(newpcb->pcb_current_pmap); PCPU_SET(curthread, new); #ifdef COMPAT_IA32 ia32_restorectx(newpcb); #endif if (PCPU_GET(fpcurthread) == new) new->td_frame->tf_special.psr &= ~IA64_PSR_DFH; restorectx(newpcb); /* We should not get here. */ panic("cpu_switch: restorectx() returned"); /* NOTREACHED */ } } void cpu_throw(struct thread *old __unused, struct thread *new) { struct pcb *newpcb; newpcb = new->td_pcb; (void)pmap_switch(newpcb->pcb_current_pmap); PCPU_SET(curthread, new); #ifdef COMPAT_IA32 ia32_restorectx(newpcb); #endif restorectx(newpcb); /* We should not get here. */ panic("cpu_throw: restorectx() returned"); /* NOTREACHED */ } void cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size) { /* * Set pc_acpi_id to "uninitialized". * See sys/dev/acpica/acpi_cpu.c */ pcpu->pc_acpi_id = 0xffffffff; } void spinlock_enter(void) { struct thread *td; td = curthread; if (td->td_md.md_spinlock_count == 0) td->td_md.md_saved_intr = intr_disable(); td->td_md.md_spinlock_count++; critical_enter(); } void spinlock_exit(void) { struct thread *td; td = curthread; critical_exit(); td->td_md.md_spinlock_count--; if (td->td_md.md_spinlock_count == 0) intr_restore(td->td_md.md_saved_intr); } void map_vhpt(uintptr_t vhpt) { pt_entry_t pte; uint64_t psr; pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY | PTE_PL_KERN | PTE_AR_RW; pte |= vhpt & PTE_PPN_MASK; __asm __volatile("ptr.d %0,%1" :: "r"(vhpt), "r"(IA64_ID_PAGE_SHIFT<<2)); __asm __volatile("mov %0=psr" : "=r"(psr)); __asm __volatile("rsm psr.ic|psr.i"); ia64_srlz_i(); ia64_set_ifa(vhpt); ia64_set_itir(IA64_ID_PAGE_SHIFT << 2); ia64_srlz_d(); __asm __volatile("itr.d dtr[%0]=%1" :: "r"(2), "r"(pte)); __asm __volatile("mov psr.l=%0" :: "r" (psr)); ia64_srlz_i(); } void map_pal_code(void) { pt_entry_t pte; uint64_t psr; if (ia64_pal_base == 0) return; pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY | PTE_PL_KERN | PTE_AR_RWX; pte |= ia64_pal_base & PTE_PPN_MASK; __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" :: "r"(IA64_PHYS_TO_RR7(ia64_pal_base)), "r"(IA64_ID_PAGE_SHIFT<<2)); __asm __volatile("mov %0=psr" : "=r"(psr)); __asm __volatile("rsm psr.ic|psr.i"); ia64_srlz_i(); ia64_set_ifa(IA64_PHYS_TO_RR7(ia64_pal_base)); ia64_set_itir(IA64_ID_PAGE_SHIFT << 2); ia64_srlz_d(); __asm __volatile("itr.d dtr[%0]=%1" :: "r"(1), "r"(pte)); ia64_srlz_d(); __asm __volatile("itr.i itr[%0]=%1" :: "r"(1), "r"(pte)); __asm __volatile("mov psr.l=%0" :: "r" (psr)); ia64_srlz_i(); } void map_gateway_page(void) { pt_entry_t pte; uint64_t psr; pte = PTE_PRESENT | PTE_MA_WB | PTE_ACCESSED | PTE_DIRTY | PTE_PL_KERN | PTE_AR_X_RX; pte |= (uint64_t)ia64_gateway_page & PTE_PPN_MASK; __asm __volatile("ptr.d %0,%1; ptr.i %0,%1" :: "r"(VM_MAX_ADDRESS), "r"(PAGE_SHIFT << 2)); __asm __volatile("mov %0=psr" : "=r"(psr)); __asm __volatile("rsm psr.ic|psr.i"); ia64_srlz_i(); ia64_set_ifa(VM_MAX_ADDRESS); ia64_set_itir(PAGE_SHIFT << 2); ia64_srlz_d(); __asm __volatile("itr.d dtr[%0]=%1" :: "r"(3), "r"(pte)); ia64_srlz_d(); __asm __volatile("itr.i itr[%0]=%1" :: "r"(3), "r"(pte)); __asm __volatile("mov psr.l=%0" :: "r" (psr)); ia64_srlz_i(); /* Expose the mapping to userland in ar.k5 */ ia64_set_k5(VM_MAX_ADDRESS); } static void calculate_frequencies(void) { struct ia64_sal_result sal; struct ia64_pal_result pal; sal = ia64_sal_entry(SAL_FREQ_BASE, 0, 0, 0, 0, 0, 0, 0); pal = ia64_call_pal_static(PAL_FREQ_RATIOS, 0, 0, 0); if (sal.sal_status == 0 && pal.pal_status == 0) { if (bootverbose) { printf("Platform clock frequency %ld Hz\n", sal.sal_result[0]); printf("Processor ratio %ld/%ld, Bus ratio %ld/%ld, " "ITC ratio %ld/%ld\n", pal.pal_result[0] >> 32, pal.pal_result[0] & ((1L << 32) - 1), pal.pal_result[1] >> 32, pal.pal_result[1] & ((1L << 32) - 1), pal.pal_result[2] >> 32, pal.pal_result[2] & ((1L << 32) - 1)); } processor_frequency = sal.sal_result[0] * (pal.pal_result[0] >> 32) / (pal.pal_result[0] & ((1L << 32) - 1)); bus_frequency = sal.sal_result[0] * (pal.pal_result[1] >> 32) / (pal.pal_result[1] & ((1L << 32) - 1)); itc_frequency = sal.sal_result[0] * (pal.pal_result[2] >> 32) / (pal.pal_result[2] & ((1L << 32) - 1)); } } struct ia64_init_return ia64_init(void) { struct ia64_init_return ret; int phys_avail_cnt; vm_offset_t kernstart, kernend; vm_offset_t kernstartpfn, kernendpfn, pfn0, pfn1; char *p; struct efi_md *md; int metadata_missing; /* NO OUTPUT ALLOWED UNTIL FURTHER NOTICE */ /* * TODO: Disable interrupts, floating point etc. * Maybe flush cache and tlb */ ia64_set_fpsr(IA64_FPSR_DEFAULT); /* * TODO: Get critical system information (if possible, from the * information provided by the boot program). */ /* * pa_bootinfo is the physical address of the bootinfo block as * passed to us by the loader and set in locore.s. */ bootinfo = *(struct bootinfo *)(IA64_PHYS_TO_RR7(pa_bootinfo)); if (bootinfo.bi_magic != BOOTINFO_MAGIC || bootinfo.bi_version != 1) { bzero(&bootinfo, sizeof(bootinfo)); bootinfo.bi_kernend = (vm_offset_t) round_page(_end); } /* * Look for the I/O ports first - we need them for console * probing. */ for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) { switch (md->md_type) { case EFI_MD_TYPE_IOPORT: ia64_port_base = IA64_PHYS_TO_RR6(md->md_phys); break; case EFI_MD_TYPE_PALCODE: ia64_pal_base = md->md_phys; break; } } metadata_missing = 0; if (bootinfo.bi_modulep) preload_metadata = (caddr_t)bootinfo.bi_modulep; else metadata_missing = 1; if (envmode == 0 && bootinfo.bi_envp) kern_envp = (caddr_t)bootinfo.bi_envp; else kern_envp = static_env; /* * Look at arguments passed to us and compute boothowto. */ boothowto = bootinfo.bi_boothowto; /* * Catch case of boot_verbose set in environment. */ if ((p = getenv("boot_verbose")) != NULL) { if (strcmp(p, "yes") == 0 || strcmp(p, "YES") == 0) { boothowto |= RB_VERBOSE; } freeenv(p); } if (boothowto & RB_VERBOSE) bootverbose = 1; /* * Find the beginning and end of the kernel. */ kernstart = trunc_page(kernel_text); #ifdef DDB ksym_start = bootinfo.bi_symtab; ksym_end = bootinfo.bi_esymtab; kernend = (vm_offset_t)round_page(ksym_end); #else kernend = (vm_offset_t)round_page(_end); #endif /* But if the bootstrap tells us otherwise, believe it! */ if (bootinfo.bi_kernend) kernend = round_page(bootinfo.bi_kernend); /* * Setup the PCPU data for the bootstrap processor. It is needed * by printf(). Also, since printf() has critical sections, we * need to initialize at least pc_curthread. */ pcpup = &pcpu0; ia64_set_k4((u_int64_t)pcpup); pcpu_init(pcpup, 0, sizeof(pcpu0)); dpcpu_init((void *)kernend, 0); kernend += DPCPU_SIZE; PCPU_SET(curthread, &thread0); /* * Initialize the console before we print anything out. */ cninit(); /* OUTPUT NOW ALLOWED */ if (ia64_pal_base != 0) { ia64_pal_base &= ~IA64_ID_PAGE_MASK; /* * We use a TR to map the first 256M of memory - this might * cover the palcode too. */ if (ia64_pal_base == 0) printf("PAL code mapped by the kernel's TR\n"); } else printf("PAL code not found\n"); /* * Wire things up so we can call the firmware. */ map_pal_code(); efi_boot_minimal(bootinfo.bi_systab); ia64_sal_init(); calculate_frequencies(); if (metadata_missing) printf("WARNING: loader(8) metadata is missing!\n"); /* Get FPSWA interface */ fpswa_iface = (bootinfo.bi_fpswa == 0) ? NULL : (struct fpswa_iface *)IA64_PHYS_TO_RR7(bootinfo.bi_fpswa); /* Init basic tunables, including hz */ init_param1(); p = getenv("kernelname"); if (p) { strncpy(kernelname, p, sizeof(kernelname) - 1); freeenv(p); } kernstartpfn = atop(IA64_RR_MASK(kernstart)); kernendpfn = atop(IA64_RR_MASK(kernend)); /* * Size the memory regions and load phys_avail[] with the results. */ /* * Find out how much memory is available, by looking at * the memory descriptors. */ #ifdef DEBUG_MD printf("Memory descriptor count: %d\n", mdcount); #endif phys_avail_cnt = 0; for (md = efi_md_first(); md != NULL; md = efi_md_next(md)) { #ifdef DEBUG_MD printf("MD %p: type %d pa 0x%lx cnt 0x%lx\n", md, md->md_type, md->md_phys, md->md_pages); #endif pfn0 = ia64_btop(round_page(md->md_phys)); pfn1 = ia64_btop(trunc_page(md->md_phys + md->md_pages * 4096)); if (pfn1 <= pfn0) continue; if (md->md_type != EFI_MD_TYPE_FREE) continue; /* * We have a memory descriptor that describes conventional * memory that is for general use. We must determine if the * loader has put the kernel in this region. */ physmem += (pfn1 - pfn0); if (pfn0 <= kernendpfn && kernstartpfn <= pfn1) { /* * Must compute the location of the kernel * within the segment. */ #ifdef DEBUG_MD printf("Descriptor %p contains kernel\n", mp); #endif if (pfn0 < kernstartpfn) { /* * There is a chunk before the kernel. */ #ifdef DEBUG_MD printf("Loading chunk before kernel: " "0x%lx / 0x%lx\n", pfn0, kernstartpfn); #endif phys_avail[phys_avail_cnt] = ia64_ptob(pfn0); phys_avail[phys_avail_cnt+1] = ia64_ptob(kernstartpfn); phys_avail_cnt += 2; } if (kernendpfn < pfn1) { /* * There is a chunk after the kernel. */ #ifdef DEBUG_MD printf("Loading chunk after kernel: " "0x%lx / 0x%lx\n", kernendpfn, pfn1); #endif phys_avail[phys_avail_cnt] = ia64_ptob(kernendpfn); phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1); phys_avail_cnt += 2; } } else { /* * Just load this cluster as one chunk. */ #ifdef DEBUG_MD printf("Loading descriptor %d: 0x%lx / 0x%lx\n", i, pfn0, pfn1); #endif phys_avail[phys_avail_cnt] = ia64_ptob(pfn0); phys_avail[phys_avail_cnt+1] = ia64_ptob(pfn1); phys_avail_cnt += 2; } } phys_avail[phys_avail_cnt] = 0; Maxmem = physmem; init_param2(physmem); /* * Initialize error message buffer (at end of core). */ msgbufp = (struct msgbuf *)pmap_steal_memory(MSGBUF_SIZE); msgbufinit(msgbufp, MSGBUF_SIZE); proc_linkup0(&proc0, &thread0); /* * Init mapping for kernel stack for proc 0 */ thread0.td_kstack = pmap_steal_memory(KSTACK_PAGES * PAGE_SIZE); thread0.td_kstack_pages = KSTACK_PAGES; mutex_init(); /* * Initialize the rest of proc 0's PCB. * * Set the kernel sp, reserving space for an (empty) trapframe, * and make proc0's trapframe pointer point to it for sanity. * Initialise proc0's backing store to start after u area. */ cpu_thread_alloc(&thread0); thread0.td_frame->tf_flags = FRAME_SYSCALL; thread0.td_pcb->pcb_special.sp = (u_int64_t)thread0.td_frame - 16; thread0.td_pcb->pcb_special.bspstore = thread0.td_kstack; /* * Initialize the virtual memory system. */ pmap_bootstrap(); /* * Initialize debuggers, and break into them if appropriate. */ kdb_init(); #ifdef KDB if (boothowto & RB_KDB) kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger\n"); #endif ia64_set_tpr(0); ia64_srlz_d(); ret.bspstore = thread0.td_pcb->pcb_special.bspstore; ret.sp = thread0.td_pcb->pcb_special.sp; return (ret); } void * ia64_ioport_address(u_int port) { uint64_t addr; addr = (port > 0xffff) ? IA64_PHYS_TO_RR6((uint64_t)port) : ia64_port_base | ((port & 0xfffc) << 10) | (port & 0xFFF); return ((void *)addr); } uint64_t ia64_get_hcdp(void) { return (bootinfo.bi_hcdp); } void bzero(void *buf, size_t len) { caddr_t p = buf; while (((vm_offset_t) p & (sizeof(u_long) - 1)) && len) { *p++ = 0; len--; } while (len >= sizeof(u_long) * 8) { *(u_long*) p = 0; *((u_long*) p + 1) = 0; *((u_long*) p + 2) = 0; *((u_long*) p + 3) = 0; len -= sizeof(u_long) * 8; *((u_long*) p + 4) = 0; *((u_long*) p + 5) = 0; *((u_long*) p + 6) = 0; *((u_long*) p + 7) = 0; p += sizeof(u_long) * 8; } while (len >= sizeof(u_long)) { *(u_long*) p = 0; len -= sizeof(u_long); p += sizeof(u_long); } while (len) { *p++ = 0; len--; } } void DELAY(int n) { u_int64_t start, end, now; sched_pin(); start = ia64_get_itc(); end = start + (itc_frequency * n) / 1000000; /* printf("DELAY from 0x%lx to 0x%lx\n", start, end); */ do { now = ia64_get_itc(); } while (now < end || (now > start && end < start)); sched_unpin(); } /* * Send an interrupt (signal) to a process. */ void sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct proc *p; struct thread *td; struct trapframe *tf; struct sigacts *psp; struct sigframe sf, *sfp; u_int64_t sbs, sp; int oonstack; int sig; u_long code; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = ksi->ksi_signo; code = ksi->ksi_code; psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); tf = td->td_frame; sp = tf->tf_special.sp; oonstack = sigonstack(sp); sbs = 0; /* save user context */ bzero(&sf, sizeof(struct sigframe)); sf.sf_uc.uc_sigmask = *mask; sf.sf_uc.uc_stack = td->td_sigstk; sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE; /* * Allocate and validate space for the signal handler * context. Note that if the stack is in P0 space, the * call to grow() is a nop, and the useracc() check * will fail if the process has not already allocated * the space with a `brk'. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { sbs = (u_int64_t)td->td_sigstk.ss_sp; sbs = (sbs + 15) & ~15; sfp = (struct sigframe *)(sbs + td->td_sigstk.ss_size); #if defined(COMPAT_43) td->td_sigstk.ss_flags |= SS_ONSTACK; #endif } else sfp = (struct sigframe *)sp; sfp = (struct sigframe *)((u_int64_t)(sfp - 1) & ~15); /* Fill in the siginfo structure for POSIX handlers. */ if (SIGISMEMBER(psp->ps_siginfo, sig)) { sf.sf_si = ksi->ksi_info; sf.sf_si.si_signo = sig; /* * XXX this shouldn't be here after code in trap.c * is fixed */ sf.sf_si.si_addr = (void*)tf->tf_special.ifa; code = (u_int64_t)&sfp->sf_si; } mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); get_mcontext(td, &sf.sf_uc.uc_mcontext, 0); /* Copy the frame out to userland. */ if (copyout(&sf, sfp, sizeof(sf)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ PROC_LOCK(p); sigexit(td, SIGILL); return; } if ((tf->tf_flags & FRAME_SYSCALL) == 0) { tf->tf_special.psr &= ~IA64_PSR_RI; tf->tf_special.iip = ia64_get_k5() + ((uint64_t)break_sigtramp - (uint64_t)ia64_gateway_page); } else tf->tf_special.iip = ia64_get_k5() + ((uint64_t)epc_sigtramp - (uint64_t)ia64_gateway_page); /* * Setup the trapframe to return to the signal trampoline. We pass * information to the trampoline in the following registers: * * gp new backing store or NULL * r8 signal number * r9 signal code or siginfo pointer * r10 signal handler (function descriptor) */ tf->tf_special.sp = (u_int64_t)sfp - 16; tf->tf_special.gp = sbs; tf->tf_special.bspstore = sf.sf_uc.uc_mcontext.mc_special.bspstore; tf->tf_special.ndirty = 0; tf->tf_special.rnat = sf.sf_uc.uc_mcontext.mc_special.rnat; tf->tf_scratch.gr8 = sig; tf->tf_scratch.gr9 = code; tf->tf_scratch.gr10 = (u_int64_t)catcher; PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by sendsig (above). * Return to previous pc and psl as specified by * context left by sendsig. Check carefully to * make sure that the user has not modified the * state to gain improper privileges. * * MPSAFE */ int sigreturn(struct thread *td, struct sigreturn_args /* { ucontext_t *sigcntxp; } */ *uap) { ucontext_t uc; struct trapframe *tf; struct pcb *pcb; tf = td->td_frame; pcb = td->td_pcb; /* * Fetch the entire context structure at once for speed. * We don't use a normal argument to simplify RSE handling. */ if (copyin(uap->sigcntxp, (caddr_t)&uc, sizeof(uc))) return (EFAULT); set_mcontext(td, &uc.uc_mcontext); #if defined(COMPAT_43) if (sigonstack(tf->tf_special.sp)) td->td_sigstk.ss_flags |= SS_ONSTACK; else td->td_sigstk.ss_flags &= ~SS_ONSTACK; #endif kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0); return (EJUSTRETURN); } #ifdef COMPAT_FREEBSD4 int freebsd4_sigreturn(struct thread *td, struct freebsd4_sigreturn_args *uap) { return sigreturn(td, (struct sigreturn_args *)uap); } #endif /* * Construct a PCB from a trapframe. This is called from kdb_trap() where * we want to start a backtrace from the function that caused us to enter * the debugger. We have the context in the trapframe, but base the trace * on the PCB. The PCB doesn't have to be perfect, as long as it contains * enough for a backtrace. */ void makectx(struct trapframe *tf, struct pcb *pcb) { pcb->pcb_special = tf->tf_special; pcb->pcb_special.__spare = ~0UL; /* XXX see unwind.c */ save_callee_saved(&pcb->pcb_preserved); save_callee_saved_fp(&pcb->pcb_preserved_fp); } int ia64_flush_dirty(struct thread *td, struct _special *r) { struct iovec iov; struct uio uio; uint64_t bspst, kstk, rnat; int error, locked; if (r->ndirty == 0) return (0); kstk = td->td_kstack + (r->bspstore & 0x1ffUL); if (td == curthread) { __asm __volatile("mov ar.rsc=0;;"); __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst)); /* Make sure we have all the user registers written out. */ if (bspst - kstk < r->ndirty) { __asm __volatile("flushrs;;"); __asm __volatile("mov %0=ar.bspstore" : "=r"(bspst)); } __asm __volatile("mov %0=ar.rnat;;" : "=r"(rnat)); __asm __volatile("mov ar.rsc=3"); error = copyout((void*)kstk, (void*)r->bspstore, r->ndirty); kstk += r->ndirty; r->rnat = (bspst > kstk && (bspst & 0x1ffL) < (kstk & 0x1ffL)) ? *(uint64_t*)(kstk | 0x1f8L) : rnat; } else { locked = PROC_LOCKED(td->td_proc); if (!locked) PHOLD(td->td_proc); iov.iov_base = (void*)(uintptr_t)kstk; iov.iov_len = r->ndirty; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = r->bspstore; uio.uio_resid = r->ndirty; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_WRITE; uio.uio_td = td; error = proc_rwmem(td->td_proc, &uio); /* * XXX proc_rwmem() doesn't currently return ENOSPC, * so I think it can bogusly return 0. Neither do * we allow short writes. */ if (uio.uio_resid != 0 && error == 0) error = ENOSPC; if (!locked) PRELE(td->td_proc); } r->bspstore += r->ndirty; r->ndirty = 0; return (error); } int get_mcontext(struct thread *td, mcontext_t *mc, int flags) { struct trapframe *tf; int error; tf = td->td_frame; bzero(mc, sizeof(*mc)); mc->mc_special = tf->tf_special; error = ia64_flush_dirty(td, &mc->mc_special); if (tf->tf_flags & FRAME_SYSCALL) { mc->mc_flags |= _MC_FLAGS_SYSCALL_CONTEXT; mc->mc_scratch = tf->tf_scratch; if (flags & GET_MC_CLEAR_RET) { mc->mc_scratch.gr8 = 0; mc->mc_scratch.gr9 = 0; mc->mc_scratch.gr10 = 0; mc->mc_scratch.gr11 = 0; } } else { mc->mc_flags |= _MC_FLAGS_ASYNC_CONTEXT; mc->mc_scratch = tf->tf_scratch; mc->mc_scratch_fp = tf->tf_scratch_fp; /* * XXX If the thread never used the high FP registers, we * probably shouldn't waste time saving them. */ ia64_highfp_save(td); mc->mc_flags |= _MC_FLAGS_HIGHFP_VALID; mc->mc_high_fp = td->td_pcb->pcb_high_fp; } save_callee_saved(&mc->mc_preserved); save_callee_saved_fp(&mc->mc_preserved_fp); return (error); } int set_mcontext(struct thread *td, const mcontext_t *mc) { struct _special s; struct trapframe *tf; uint64_t psrmask; tf = td->td_frame; KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0, ("Whoa there! We have more than 8KB of dirty registers!")); s = mc->mc_special; /* * Only copy the user mask and the restart instruction bit from * the new context. */ psrmask = IA64_PSR_BE | IA64_PSR_UP | IA64_PSR_AC | IA64_PSR_MFL | IA64_PSR_MFH | IA64_PSR_RI; s.psr = (tf->tf_special.psr & ~psrmask) | (s.psr & psrmask); /* We don't have any dirty registers of the new context. */ s.ndirty = 0; if (mc->mc_flags & _MC_FLAGS_ASYNC_CONTEXT) { /* * We can get an async context passed to us while we * entered the kernel through a syscall: sigreturn(2) * takes contexts that could previously be the result of * a trap or interrupt. * Hence, we cannot assert that the trapframe is not * a syscall frame, but we can assert that it's at * least an expected syscall. */ if (tf->tf_flags & FRAME_SYSCALL) { KASSERT(tf->tf_scratch.gr15 == SYS_sigreturn, ("foo")); tf->tf_flags &= ~FRAME_SYSCALL; } tf->tf_scratch = mc->mc_scratch; tf->tf_scratch_fp = mc->mc_scratch_fp; if (mc->mc_flags & _MC_FLAGS_HIGHFP_VALID) td->td_pcb->pcb_high_fp = mc->mc_high_fp; } else { KASSERT((tf->tf_flags & FRAME_SYSCALL) != 0, ("foo")); if ((mc->mc_flags & _MC_FLAGS_SYSCALL_CONTEXT) == 0) { s.cfm = tf->tf_special.cfm; s.iip = tf->tf_special.iip; tf->tf_scratch.gr15 = 0; /* Clear syscall nr. */ } else tf->tf_scratch = mc->mc_scratch; } tf->tf_special = s; restore_callee_saved(&mc->mc_preserved); restore_callee_saved_fp(&mc->mc_preserved_fp); return (0); } /* * Clear registers on exec. */ void exec_setregs(struct thread *td, u_long entry, u_long stack, u_long ps_strings) { struct trapframe *tf; uint64_t *ksttop, *kst; tf = td->td_frame; ksttop = (uint64_t*)(td->td_kstack + tf->tf_special.ndirty + (tf->tf_special.bspstore & 0x1ffUL)); /* * We can ignore up to 8KB of dirty registers by masking off the * lower 13 bits in exception_restore() or epc_syscall(). This * should be enough for a couple of years, but if there are more * than 8KB of dirty registers, we lose track of the bottom of * the kernel stack. The solution is to copy the active part of * the kernel stack down 1 page (or 2, but not more than that) * so that we always have less than 8KB of dirty registers. */ KASSERT((tf->tf_special.ndirty & ~PAGE_MASK) == 0, ("Whoa there! We have more than 8KB of dirty registers!")); bzero(&tf->tf_special, sizeof(tf->tf_special)); if ((tf->tf_flags & FRAME_SYSCALL) == 0) { /* break syscalls. */ bzero(&tf->tf_scratch, sizeof(tf->tf_scratch)); bzero(&tf->tf_scratch_fp, sizeof(tf->tf_scratch_fp)); tf->tf_special.cfm = (1UL<<63) | (3UL<<7) | 3UL; tf->tf_special.bspstore = IA64_BACKINGSTORE; /* * Copy the arguments onto the kernel register stack so that * they get loaded by the loadrs instruction. Skip over the * NaT collection points. */ kst = ksttop - 1; if (((uintptr_t)kst & 0x1ff) == 0x1f8) *kst-- = 0; *kst-- = 0; if (((uintptr_t)kst & 0x1ff) == 0x1f8) *kst-- = 0; *kst-- = ps_strings; if (((uintptr_t)kst & 0x1ff) == 0x1f8) *kst-- = 0; *kst = stack; tf->tf_special.ndirty = (ksttop - kst) << 3; } else { /* epc syscalls (default). */ tf->tf_special.cfm = (3UL<<62) | (3UL<<7) | 3UL; tf->tf_special.bspstore = IA64_BACKINGSTORE + 24; /* * Write values for out0, out1 and out2 to the user's backing * store and arrange for them to be restored into the user's * initial register frame. * Assumes that (bspstore & 0x1f8) < 0x1e0. */ suword((caddr_t)tf->tf_special.bspstore - 24, stack); suword((caddr_t)tf->tf_special.bspstore - 16, ps_strings); suword((caddr_t)tf->tf_special.bspstore - 8, 0); } tf->tf_special.iip = entry; tf->tf_special.sp = (stack & ~15) - 16; tf->tf_special.rsc = 0xf; tf->tf_special.fpsr = IA64_FPSR_DEFAULT; tf->tf_special.psr = IA64_PSR_IC | IA64_PSR_I | IA64_PSR_IT | IA64_PSR_DT | IA64_PSR_RT | IA64_PSR_DFH | IA64_PSR_BN | IA64_PSR_CPL_USER; } int ptrace_set_pc(struct thread *td, unsigned long addr) { uint64_t slot; switch (addr & 0xFUL) { case 0: slot = IA64_PSR_RI_0; break; case 1: /* XXX we need to deal with MLX bundles here */ slot = IA64_PSR_RI_1; break; case 2: slot = IA64_PSR_RI_2; break; default: return (EINVAL); } td->td_frame->tf_special.iip = addr & ~0x0FULL; td->td_frame->tf_special.psr = (td->td_frame->tf_special.psr & ~IA64_PSR_RI) | slot; return (0); } int ptrace_single_step(struct thread *td) { struct trapframe *tf; /* * There's no way to set single stepping when we're leaving the * kernel through the EPC syscall path. The way we solve this is * by enabling the lower-privilege trap so that we re-enter the * kernel as soon as the privilege level changes. See trap.c for * how we proceed from there. */ tf = td->td_frame; if (tf->tf_flags & FRAME_SYSCALL) tf->tf_special.psr |= IA64_PSR_LP; else tf->tf_special.psr |= IA64_PSR_SS; return (0); } int ptrace_clear_single_step(struct thread *td) { struct trapframe *tf; /* * Clear any and all status bits we may use to implement single * stepping. */ tf = td->td_frame; tf->tf_special.psr &= ~IA64_PSR_SS; tf->tf_special.psr &= ~IA64_PSR_LP; tf->tf_special.psr &= ~IA64_PSR_TB; return (0); } int fill_regs(struct thread *td, struct reg *regs) { struct trapframe *tf; tf = td->td_frame; regs->r_special = tf->tf_special; regs->r_scratch = tf->tf_scratch; save_callee_saved(®s->r_preserved); return (0); } int set_regs(struct thread *td, struct reg *regs) { struct trapframe *tf; int error; tf = td->td_frame; error = ia64_flush_dirty(td, &tf->tf_special); if (!error) { tf->tf_special = regs->r_special; tf->tf_special.bspstore += tf->tf_special.ndirty; tf->tf_special.ndirty = 0; tf->tf_scratch = regs->r_scratch; restore_callee_saved(®s->r_preserved); } return (error); } int fill_dbregs(struct thread *td, struct dbreg *dbregs) { return (ENOSYS); } int set_dbregs(struct thread *td, struct dbreg *dbregs) { return (ENOSYS); } int fill_fpregs(struct thread *td, struct fpreg *fpregs) { struct trapframe *frame = td->td_frame; struct pcb *pcb = td->td_pcb; /* Save the high FP registers. */ ia64_highfp_save(td); fpregs->fpr_scratch = frame->tf_scratch_fp; save_callee_saved_fp(&fpregs->fpr_preserved); fpregs->fpr_high = pcb->pcb_high_fp; return (0); } int set_fpregs(struct thread *td, struct fpreg *fpregs) { struct trapframe *frame = td->td_frame; struct pcb *pcb = td->td_pcb; /* Throw away the high FP registers (should be redundant). */ ia64_highfp_drop(td); frame->tf_scratch_fp = fpregs->fpr_scratch; restore_callee_saved_fp(&fpregs->fpr_preserved); pcb->pcb_high_fp = fpregs->fpr_high; return (0); } void ia64_sync_icache(vm_offset_t va, vm_offset_t sz) { vm_offset_t lim; if (!ia64_sync_icache_needed) return; lim = va + sz; while (va < lim) { ia64_fc_i(va); va += 32; /* XXX */ } ia64_sync_i(); ia64_srlz_i(); } diff --git a/sys/ia64/ia64/mp_machdep.c b/sys/ia64/ia64/mp_machdep.c index eecf6ac08c25..020c71b0f836 100644 --- a/sys/ia64/ia64/mp_machdep.c +++ b/sys/ia64/ia64/mp_machdep.c @@ -1,378 +1,378 @@ /*- * Copyright (c) 2001-2005 Marcel Moolenaar * Copyright (c) 2000 Doug Rabson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_kstack_pages.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_SMP, "SMP", "SMP related allocations"); void ia64_ap_startup(void); #define LID_SAPIC_ID(x) ((int)((x) >> 24) & 0xff) #define LID_SAPIC_EID(x) ((int)((x) >> 16) & 0xff) #define LID_SAPIC_SET(id,eid) (((id & 0xff) << 8 | (eid & 0xff)) << 16); #define LID_SAPIC_MASK 0xffff0000UL /* Variables used by os_boot_rendez and ia64_ap_startup */ struct pcpu *ap_pcpu; void *ap_stack; volatile int ap_delay; volatile int ap_awake; volatile int ap_spin; static void cpu_mp_unleash(void *); struct cpu_group * cpu_topo(void) { return smp_topo_none(); } static void ia64_store_mca_state(void* arg) { unsigned int ncpu = (unsigned int)(uintptr_t)arg; struct thread* td; /* ia64_mca_save_state() is CPU-sensitive, so bind ourself to our target CPU */ td = curthread; thread_lock(td); sched_bind(td, ncpu); thread_unlock(td); /* * Get and save the CPU specific MCA records. Should we get the * MCA state for each processor, or just the CMC state? */ ia64_mca_save_state(SAL_INFO_MCA); ia64_mca_save_state(SAL_INFO_CMC); kproc_exit(0); } void ia64_ap_startup(void) { volatile struct ia64_interrupt_block *ib = IA64_INTERRUPT_BLOCK; uint64_t vhpt; int vector; pcpup = ap_pcpu; ia64_set_k4((intptr_t)pcpup); - vhpt = PCPU_GET(vhpt); + vhpt = PCPU_GET(md.vhpt); map_vhpt(vhpt); ia64_set_pta(vhpt + (1 << 8) + (pmap_vhpt_log2size << 2) + 1); ia64_srlz_i(); ap_awake = 1; ap_delay = 0; map_pal_code(); map_gateway_page(); ia64_set_fpsr(IA64_FPSR_DEFAULT); /* Wait until it's time for us to be unleashed */ while (ap_spin) cpu_spinwait(); /* Initialize curthread. */ KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread")); PCPU_SET(curthread, PCPU_GET(idlethread)); atomic_add_int(&ap_awake, 1); while (!smp_started) cpu_spinwait(); CTR1(KTR_SMP, "SMP: cpu%d launched", PCPU_GET(cpuid)); /* Acknowledge and EOI all interrupts. */ vector = ia64_get_ivr(); while (vector != 15) { ia64_srlz_d(); if (vector == 0) vector = (int)ib->ib_inta; ia64_set_eoi(0); ia64_srlz_d(); vector = ia64_get_ivr(); } ia64_srlz_d(); /* kick off the clock on this AP */ pcpu_initclock(); ia64_set_tpr(0); ia64_srlz_d(); enable_intr(); sched_throw(NULL); /* NOTREACHED */ } void cpu_mp_setmaxid(void) { /* * Count the number of processors in the system by walking the ACPI * tables. Note that we record the actual number of processors, even * if this is larger than MAXCPU. We only activate MAXCPU processors. */ mp_ncpus = ia64_count_cpus(); /* * Set the largest cpuid we're going to use. This is necessary for * VM initialization. */ mp_maxid = min(mp_ncpus, MAXCPU) - 1; } int cpu_mp_probe(void) { /* * If there's only 1 processor, or we don't have a wake-up vector, * we're not going to enable SMP. Note that no wake-up vector can * also mean that the wake-up mechanism is not supported. In this * case we can have multiple processors, but we simply can't wake * them up... */ return (mp_ncpus > 1 && ipi_vector[IPI_AP_WAKEUP] != 0); } void cpu_mp_add(u_int acpiid, u_int apicid, u_int apiceid) { struct pcpu *pc; u_int64_t lid; void *dpcpu; u_int cpuid; lid = LID_SAPIC_SET(apicid, apiceid); cpuid = ((ia64_get_lid() & LID_SAPIC_MASK) == lid) ? 0 : smp_cpus++; KASSERT((all_cpus & (1UL << cpuid)) == 0, ("%s: cpu%d already in CPU map", __func__, acpiid)); if (cpuid != 0) { pc = (struct pcpu *)malloc(sizeof(*pc), M_SMP, M_WAITOK); pcpu_init(pc, cpuid, sizeof(*pc)); dpcpu = (void *)kmem_alloc(kernel_map, DPCPU_SIZE); dpcpu_init(dpcpu, cpuid); } else pc = pcpup; pc->pc_acpi_id = acpiid; - pc->pc_lid = lid; + pc->pc_md.lid = lid; all_cpus |= (1UL << cpuid); } void cpu_mp_announce() { struct pcpu *pc; int i; for (i = 0; i <= mp_maxid; i++) { pc = pcpu_find(i); if (pc != NULL) { printf("cpu%d: ACPI Id=%x, SAPIC Id=%x, SAPIC Eid=%x", - i, pc->pc_acpi_id, LID_SAPIC_ID(pc->pc_lid), - LID_SAPIC_EID(pc->pc_lid)); + i, pc->pc_acpi_id, LID_SAPIC_ID(pc->pc_md.lid), + LID_SAPIC_EID(pc->pc_md.lid)); if (i == 0) printf(" (BSP)\n"); else printf("\n"); } } } void cpu_mp_start() { struct pcpu *pc; ap_spin = 1; SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { - pc->pc_current_pmap = kernel_pmap; + pc->pc_md.current_pmap = kernel_pmap; pc->pc_other_cpus = all_cpus & ~pc->pc_cpumask; if (pc->pc_cpuid > 0) { ap_pcpu = pc; - pc->pc_vhpt = pmap_alloc_vhpt(); - if (pc->pc_vhpt == 0) { + pc->pc_md.vhpt = pmap_alloc_vhpt(); + if (pc->pc_md.vhpt == 0) { printf("SMP: WARNING: unable to allocate VHPT" " for cpu%d", pc->pc_cpuid); continue; } ap_stack = malloc(KSTACK_PAGES * PAGE_SIZE, M_SMP, M_WAITOK); ap_delay = 2000; ap_awake = 0; if (bootverbose) printf("SMP: waking up cpu%d\n", pc->pc_cpuid); ipi_send(pc, IPI_AP_WAKEUP); do { DELAY(1000); } while (--ap_delay > 0); - pc->pc_awake = ap_awake; + pc->pc_md.awake = ap_awake; if (!ap_awake) printf("SMP: WARNING: cpu%d did not wake up\n", pc->pc_cpuid); } else - pc->pc_awake = 1; + pc->pc_md.awake = 1; } } static void cpu_mp_unleash(void *dummy) { struct pcpu *pc; int cpus; if (mp_ncpus <= 1) return; cpus = 0; smp_cpus = 0; SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { cpus++; - if (pc->pc_awake) { + if (pc->pc_md.awake) { kproc_create(ia64_store_mca_state, (void*)((uintptr_t)pc->pc_cpuid), NULL, 0, 0, "mca %u", pc->pc_cpuid); smp_cpus++; } } ap_awake = 1; ap_spin = 0; while (ap_awake != smp_cpus) cpu_spinwait(); if (smp_cpus != cpus || cpus != mp_ncpus) { printf("SMP: %d CPUs found; %d CPUs usable; %d CPUs woken\n", mp_ncpus, cpus, smp_cpus); } smp_active = 1; smp_started = 1; } /* * send an IPI to a set of cpus. */ void ipi_selected(cpumask_t cpus, int ipi) { struct pcpu *pc; SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { if (cpus & pc->pc_cpumask) ipi_send(pc, ipi); } } /* * send an IPI to all CPUs EXCEPT myself. */ void ipi_all_but_self(int ipi) { struct pcpu *pc; SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { if (pc != pcpup) ipi_send(pc, ipi); } } /* * Send an IPI to the specified processor. The lid parameter holds the * cr.lid (CR64) contents of the target processor. Only the id and eid * fields are used here. */ void ipi_send(struct pcpu *cpu, int ipi) { volatile uint64_t *pipi; uint64_t vector; pipi = __MEMIO_ADDR(ia64_lapic_address | - ((cpu->pc_lid & LID_SAPIC_MASK) >> 12)); + ((cpu->pc_md.lid & LID_SAPIC_MASK) >> 12)); vector = (uint64_t)(ipi_vector[ipi] & 0xff); KASSERT(vector != 0, ("IPI %d is not assigned a vector", ipi)); *pipi = vector; CTR3(KTR_SMP, "ipi_send(%p, %ld), cpuid=%d", pipi, vector, PCPU_GET(cpuid)); } SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, cpu_mp_unleash, NULL); diff --git a/sys/ia64/ia64/pmap.c b/sys/ia64/ia64/pmap.c index 315e7e543149..6a0e21498443 100644 --- a/sys/ia64/ia64/pmap.c +++ b/sys/ia64/ia64/pmap.c @@ -1,2441 +1,2441 @@ /*- * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1994 John S. Dyson * All rights reserved. * Copyright (c) 1994 David Greenman * All rights reserved. * Copyright (c) 1998,2000 Doug Rabson * All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department and William Jolitz of UUNET Technologies Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91 * from: i386 Id: pmap.c,v 1.193 1998/04/19 15:22:48 bde Exp * with some ideas from NetBSD's alpha pmap */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Manages physical address maps. * * In addition to hardware address maps, this * module is called upon to provide software-use-only * maps which may or may not be stored in the same * form as hardware maps. These pseudo-maps are * used to store intermediate results from copy * operations to and from address spaces. * * Since the information managed by this module is * also stored by the logical address mapping module, * this module may throw away valid virtual-to-physical * mappings at almost any time. However, invalidations * of virtual-to-physical mappings must be done as * requested. * * In order to cope with hardware architectures which * make virtual-to-physical map invalidates expensive, * this module may delay invalidate or reduced protection * operations until such time as they are actually * necessary. This module is given full information as * to which processors are currently using which maps, * and to when physical maps must be made correct. */ /* * Following the Linux model, region IDs are allocated in groups of * eight so that a single region ID can be used for as many RRs as we * want by encoding the RR number into the low bits of the ID. * * We reserve region ID 0 for the kernel and allocate the remaining * IDs for user pmaps. * * Region 0..4 * User virtually mapped * * Region 5 * Kernel virtually mapped * * Region 6 * Kernel physically mapped uncacheable * * Region 7 * Kernel physically mapped cacheable */ /* XXX move to a header. */ extern uint64_t ia64_gateway_page[]; #ifndef PMAP_SHPGPERPROC #define PMAP_SHPGPERPROC 200 #endif #if !defined(DIAGNOSTIC) #define PMAP_INLINE __inline #else #define PMAP_INLINE #endif #define pmap_accessed(lpte) ((lpte)->pte & PTE_ACCESSED) #define pmap_dirty(lpte) ((lpte)->pte & PTE_DIRTY) #define pmap_exec(lpte) ((lpte)->pte & PTE_AR_RX) #define pmap_managed(lpte) ((lpte)->pte & PTE_MANAGED) #define pmap_ppn(lpte) ((lpte)->pte & PTE_PPN_MASK) #define pmap_present(lpte) ((lpte)->pte & PTE_PRESENT) #define pmap_prot(lpte) (((lpte)->pte & PTE_PROT_MASK) >> 56) #define pmap_wired(lpte) ((lpte)->pte & PTE_WIRED) #define pmap_clear_accessed(lpte) (lpte)->pte &= ~PTE_ACCESSED #define pmap_clear_dirty(lpte) (lpte)->pte &= ~PTE_DIRTY #define pmap_clear_present(lpte) (lpte)->pte &= ~PTE_PRESENT #define pmap_clear_wired(lpte) (lpte)->pte &= ~PTE_WIRED #define pmap_set_wired(lpte) (lpte)->pte |= PTE_WIRED /* * The VHPT bucket head structure. */ struct ia64_bucket { uint64_t chain; struct mtx mutex; u_int length; }; /* * Statically allocated kernel pmap */ struct pmap kernel_pmap_store; vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ /* * Kernel virtual memory management. */ static int nkpt; struct ia64_lpte ***ia64_kptdir; #define KPTE_DIR0_INDEX(va) \ (((va) >> (3*PAGE_SHIFT-8)) & ((1<<(PAGE_SHIFT-3))-1)) #define KPTE_DIR1_INDEX(va) \ (((va) >> (2*PAGE_SHIFT-5)) & ((1<<(PAGE_SHIFT-3))-1)) #define KPTE_PTE_INDEX(va) \ (((va) >> PAGE_SHIFT) & ((1<<(PAGE_SHIFT-5))-1)) #define NKPTEPG (PAGE_SIZE / sizeof(struct ia64_lpte)) vm_offset_t kernel_vm_end; /* Values for ptc.e. XXX values for SKI. */ static uint64_t pmap_ptc_e_base = 0x100000000; static uint64_t pmap_ptc_e_count1 = 3; static uint64_t pmap_ptc_e_count2 = 2; static uint64_t pmap_ptc_e_stride1 = 0x2000; static uint64_t pmap_ptc_e_stride2 = 0x100000000; struct mtx pmap_ptcmutex; /* * Data for the RID allocator */ static int pmap_ridcount; static int pmap_rididx; static int pmap_ridmapsz; static int pmap_ridmax; static uint64_t *pmap_ridmap; struct mtx pmap_ridmutex; /* * Data for the pv entry allocation mechanism */ static uma_zone_t pvzone; static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0; /* * Data for allocating PTEs for user processes. */ static uma_zone_t ptezone; /* * Virtual Hash Page Table (VHPT) data. */ /* SYSCTL_DECL(_machdep); */ SYSCTL_NODE(_machdep, OID_AUTO, vhpt, CTLFLAG_RD, 0, ""); struct ia64_bucket *pmap_vhpt_bucket; int pmap_vhpt_nbuckets; SYSCTL_INT(_machdep_vhpt, OID_AUTO, nbuckets, CTLFLAG_RD, &pmap_vhpt_nbuckets, 0, ""); int pmap_vhpt_log2size = 0; TUNABLE_INT("machdep.vhpt.log2size", &pmap_vhpt_log2size); SYSCTL_INT(_machdep_vhpt, OID_AUTO, log2size, CTLFLAG_RD, &pmap_vhpt_log2size, 0, ""); static int pmap_vhpt_inserts; SYSCTL_INT(_machdep_vhpt, OID_AUTO, inserts, CTLFLAG_RD, &pmap_vhpt_inserts, 0, ""); static int pmap_vhpt_population(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_machdep_vhpt, OID_AUTO, population, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, pmap_vhpt_population, "I", ""); static struct ia64_lpte *pmap_find_vhpt(vm_offset_t va); static PMAP_INLINE void free_pv_entry(pv_entry_t pv); static pv_entry_t get_pv_entry(pmap_t locked_pmap); static void pmap_enter_quick_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot); static void pmap_free_pte(struct ia64_lpte *pte, vm_offset_t va); static void pmap_invalidate_all(pmap_t pmap); static int pmap_remove_pte(pmap_t pmap, struct ia64_lpte *pte, vm_offset_t va, pv_entry_t pv, int freepte); static int pmap_remove_vhpt(vm_offset_t va); static boolean_t pmap_try_insert_pv_entry(pmap_t pmap, vm_offset_t va, vm_page_t m); vm_offset_t pmap_steal_memory(vm_size_t size) { vm_size_t bank_size; vm_offset_t pa, va; size = round_page(size); bank_size = phys_avail[1] - phys_avail[0]; while (size > bank_size) { int i; for (i = 0; phys_avail[i+2]; i+= 2) { phys_avail[i] = phys_avail[i+2]; phys_avail[i+1] = phys_avail[i+3]; } phys_avail[i] = 0; phys_avail[i+1] = 0; if (!phys_avail[0]) panic("pmap_steal_memory: out of memory"); bank_size = phys_avail[1] - phys_avail[0]; } pa = phys_avail[0]; phys_avail[0] += size; va = IA64_PHYS_TO_RR7(pa); bzero((caddr_t) va, size); return va; } static void pmap_initialize_vhpt(vm_offset_t vhpt) { struct ia64_lpte *pte; u_int i; pte = (struct ia64_lpte *)vhpt; for (i = 0; i < pmap_vhpt_nbuckets; i++) { pte[i].pte = 0; pte[i].itir = 0; pte[i].tag = 1UL << 63; /* Invalid tag */ pte[i].chain = (uintptr_t)(pmap_vhpt_bucket + i); } } #ifdef SMP MALLOC_DECLARE(M_SMP); vm_offset_t pmap_alloc_vhpt(void) { vm_offset_t vhpt; vm_size_t size; size = 1UL << pmap_vhpt_log2size; vhpt = (uintptr_t)contigmalloc(size, M_SMP, 0, 0UL, ~0UL, size, 0UL); if (vhpt != 0) { vhpt = IA64_PHYS_TO_RR7(ia64_tpa(vhpt)); pmap_initialize_vhpt(vhpt); } return (vhpt); } #endif /* * Bootstrap the system enough to run with virtual memory. */ void pmap_bootstrap() { struct ia64_pal_result res; vm_offset_t base; size_t size; int i, j, count, ridbits; /* * Query the PAL Code to find the loop parameters for the * ptc.e instruction. */ res = ia64_call_pal_static(PAL_PTCE_INFO, 0, 0, 0); if (res.pal_status != 0) panic("Can't configure ptc.e parameters"); pmap_ptc_e_base = res.pal_result[0]; pmap_ptc_e_count1 = res.pal_result[1] >> 32; pmap_ptc_e_count2 = res.pal_result[1] & ((1L<<32) - 1); pmap_ptc_e_stride1 = res.pal_result[2] >> 32; pmap_ptc_e_stride2 = res.pal_result[2] & ((1L<<32) - 1); if (bootverbose) printf("ptc.e base=0x%lx, count1=%ld, count2=%ld, " "stride1=0x%lx, stride2=0x%lx\n", pmap_ptc_e_base, pmap_ptc_e_count1, pmap_ptc_e_count2, pmap_ptc_e_stride1, pmap_ptc_e_stride2); mtx_init(&pmap_ptcmutex, "Global PTC lock", NULL, MTX_SPIN); /* * Setup RIDs. RIDs 0..7 are reserved for the kernel. * * We currently need at least 19 bits in the RID because PID_MAX * can only be encoded in 17 bits and we need RIDs for 5 regions * per process. With PID_MAX equalling 99999 this means that we * need to be able to encode 499995 (=5*PID_MAX). * The Itanium processor only has 18 bits and the architected * minimum is exactly that. So, we cannot use a PID based scheme * in those cases. Enter pmap_ridmap... * We should avoid the map when running on a processor that has * implemented enough bits. This means that we should pass the * process/thread ID to pmap. This we currently don't do, so we * use the map anyway. However, we don't want to allocate a map * that is large enough to cover the range dictated by the number * of bits in the RID, because that may result in a RID map of * 2MB in size for a 24-bit RID. A 64KB map is enough. * The bottomline: we create a 32KB map when the processor only * implements 18 bits (or when we can't figure it out). Otherwise * we create a 64KB map. */ res = ia64_call_pal_static(PAL_VM_SUMMARY, 0, 0, 0); if (res.pal_status != 0) { if (bootverbose) printf("Can't read VM Summary - assuming 18 Region ID bits\n"); ridbits = 18; /* guaranteed minimum */ } else { ridbits = (res.pal_result[1] >> 8) & 0xff; if (bootverbose) printf("Processor supports %d Region ID bits\n", ridbits); } if (ridbits > 19) ridbits = 19; pmap_ridmax = (1 << ridbits); pmap_ridmapsz = pmap_ridmax / 64; pmap_ridmap = (uint64_t *)pmap_steal_memory(pmap_ridmax / 8); pmap_ridmap[0] |= 0xff; pmap_rididx = 0; pmap_ridcount = 8; mtx_init(&pmap_ridmutex, "RID allocator lock", NULL, MTX_DEF); /* * Allocate some memory for initial kernel 'page tables'. */ ia64_kptdir = (void *)pmap_steal_memory(PAGE_SIZE); nkpt = 0; kernel_vm_end = VM_MIN_KERNEL_ADDRESS - VM_GATEWAY_SIZE; for (i = 0; phys_avail[i+2]; i+= 2) ; count = i+2; TUNABLE_INT_FETCH("machdep.vhpt.log2size", &pmap_vhpt_log2size); if (pmap_vhpt_log2size == 0) pmap_vhpt_log2size = 20; else if (pmap_vhpt_log2size < 15) pmap_vhpt_log2size = 15; else if (pmap_vhpt_log2size > 61) pmap_vhpt_log2size = 61; base = 0; size = 1UL << pmap_vhpt_log2size; for (i = 0; i < count; i += 2) { base = (phys_avail[i] + size - 1) & ~(size - 1); if (base + size <= phys_avail[i+1]) break; } if (!phys_avail[i]) panic("Unable to allocate VHPT"); if (base != phys_avail[i]) { /* Split this region. */ for (j = count; j > i; j -= 2) { phys_avail[j] = phys_avail[j-2]; phys_avail[j+1] = phys_avail[j-2+1]; } phys_avail[i+1] = base; phys_avail[i+2] = base + size; } else phys_avail[i] = base + size; base = IA64_PHYS_TO_RR7(base); - PCPU_SET(vhpt, base); + PCPU_SET(md.vhpt, base); if (bootverbose) printf("VHPT: address=%#lx, size=%#lx\n", base, size); pmap_vhpt_nbuckets = size / sizeof(struct ia64_lpte); pmap_vhpt_bucket = (void *)pmap_steal_memory(pmap_vhpt_nbuckets * sizeof(struct ia64_bucket)); for (i = 0; i < pmap_vhpt_nbuckets; i++) { /* Stolen memory is zeroed. */ mtx_init(&pmap_vhpt_bucket[i].mutex, "VHPT bucket lock", NULL, MTX_NOWITNESS | MTX_SPIN); } pmap_initialize_vhpt(base); map_vhpt(base); ia64_set_pta(base + (1 << 8) + (pmap_vhpt_log2size << 2) + 1); ia64_srlz_i(); virtual_avail = VM_MIN_KERNEL_ADDRESS; virtual_end = VM_MAX_KERNEL_ADDRESS; /* * Initialize the kernel pmap (which is statically allocated). */ PMAP_LOCK_INIT(kernel_pmap); for (i = 0; i < 5; i++) kernel_pmap->pm_rid[i] = 0; kernel_pmap->pm_active = 1; TAILQ_INIT(&kernel_pmap->pm_pvlist); - PCPU_SET(current_pmap, kernel_pmap); + PCPU_SET(md.current_pmap, kernel_pmap); /* * Region 5 is mapped via the vhpt. */ ia64_set_rr(IA64_RR_BASE(5), (5 << 8) | (PAGE_SHIFT << 2) | 1); /* * Region 6 is direct mapped UC and region 7 is direct mapped * WC. The details of this is controlled by the Alt {I,D}TLB * handlers. Here we just make sure that they have the largest * possible page size to minimise TLB usage. */ ia64_set_rr(IA64_RR_BASE(6), (6 << 8) | (IA64_ID_PAGE_SHIFT << 2)); ia64_set_rr(IA64_RR_BASE(7), (7 << 8) | (IA64_ID_PAGE_SHIFT << 2)); ia64_srlz_d(); /* * Clear out any random TLB entries left over from booting. */ pmap_invalidate_all(kernel_pmap); map_gateway_page(); } static int pmap_vhpt_population(SYSCTL_HANDLER_ARGS) { int count, error, i; count = 0; for (i = 0; i < pmap_vhpt_nbuckets; i++) count += pmap_vhpt_bucket[i].length; error = SYSCTL_OUT(req, &count, sizeof(count)); return (error); } /* * Initialize a vm_page's machine-dependent fields. */ void pmap_page_init(vm_page_t m) { TAILQ_INIT(&m->md.pv_list); m->md.pv_list_count = 0; } /* * Initialize the pmap module. * Called by vm_init, to initialize any structures that the pmap * system needs to map virtual memory. */ void pmap_init(void) { int shpgperproc = PMAP_SHPGPERPROC; /* * Initialize the address space (zone) for the pv entries. Set a * high water mark so that the system can recover from excessive * numbers of pv entries. */ pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE); TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc); pv_entry_max = shpgperproc * maxproc + cnt.v_page_count; TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max); pv_entry_high_water = 9 * (pv_entry_max / 10); ptezone = uma_zcreate("PT ENTRY", sizeof (struct ia64_lpte), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM|UMA_ZONE_NOFREE); } /*************************************************** * Manipulate TLBs for a pmap ***************************************************/ static void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { struct ia64_lpte *pte; struct pcpu *pc; u_int vhpt_ofs; - KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(current_pmap)), + KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(md.current_pmap)), ("invalidating TLB for non-current pmap")); - vhpt_ofs = ia64_thash(va) - PCPU_GET(vhpt); + vhpt_ofs = ia64_thash(va) - PCPU_GET(md.vhpt); critical_enter(); SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { - pte = (struct ia64_lpte *)(pc->pc_vhpt + vhpt_ofs); + pte = (struct ia64_lpte *)(pc->pc_md.vhpt + vhpt_ofs); if (pte->tag == ia64_ttag(va)) pte->tag = 1UL << 63; } critical_exit(); mtx_lock_spin(&pmap_ptcmutex); ia64_ptc_ga(va, PAGE_SHIFT << 2); mtx_unlock_spin(&pmap_ptcmutex); } static void pmap_invalidate_all_1(void *arg) { uint64_t addr; int i, j; critical_enter(); addr = pmap_ptc_e_base; for (i = 0; i < pmap_ptc_e_count1; i++) { for (j = 0; j < pmap_ptc_e_count2; j++) { ia64_ptc_e(addr); addr += pmap_ptc_e_stride2; } addr += pmap_ptc_e_stride1; } critical_exit(); } static void pmap_invalidate_all(pmap_t pmap) { - KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(current_pmap)), + KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(md.current_pmap)), ("invalidating TLB for non-current pmap")); #ifdef SMP if (mp_ncpus > 1) smp_rendezvous(NULL, pmap_invalidate_all_1, NULL, NULL); else #endif pmap_invalidate_all_1(NULL); } static uint32_t pmap_allocate_rid(void) { uint64_t bit, bits; int rid; mtx_lock(&pmap_ridmutex); if (pmap_ridcount == pmap_ridmax) panic("pmap_allocate_rid: All Region IDs used"); /* Find an index with a free bit. */ while ((bits = pmap_ridmap[pmap_rididx]) == ~0UL) { pmap_rididx++; if (pmap_rididx == pmap_ridmapsz) pmap_rididx = 0; } rid = pmap_rididx * 64; /* Find a free bit. */ bit = 1UL; while (bits & bit) { rid++; bit <<= 1; } pmap_ridmap[pmap_rididx] |= bit; pmap_ridcount++; mtx_unlock(&pmap_ridmutex); return rid; } static void pmap_free_rid(uint32_t rid) { uint64_t bit; int idx; idx = rid / 64; bit = ~(1UL << (rid & 63)); mtx_lock(&pmap_ridmutex); pmap_ridmap[idx] &= bit; pmap_ridcount--; mtx_unlock(&pmap_ridmutex); } /*************************************************** * Page table page management routines..... ***************************************************/ void pmap_pinit0(struct pmap *pmap) { /* kernel_pmap is the same as any other pmap. */ pmap_pinit(pmap); } /* * Initialize a preallocated and zeroed pmap structure, * such as one in a vmspace structure. */ int pmap_pinit(struct pmap *pmap) { int i; PMAP_LOCK_INIT(pmap); for (i = 0; i < 5; i++) pmap->pm_rid[i] = pmap_allocate_rid(); pmap->pm_active = 0; TAILQ_INIT(&pmap->pm_pvlist); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); return (1); } /*************************************************** * Pmap allocation/deallocation routines. ***************************************************/ /* * Release any resources held by the given physical map. * Called when a pmap initialized by pmap_pinit is being released. * Should only be called if the map contains no valid mappings. */ void pmap_release(pmap_t pmap) { int i; for (i = 0; i < 5; i++) if (pmap->pm_rid[i]) pmap_free_rid(pmap->pm_rid[i]); PMAP_LOCK_DESTROY(pmap); } /* * grow the number of kernel page table entries, if needed */ void pmap_growkernel(vm_offset_t addr) { struct ia64_lpte **dir1; struct ia64_lpte *leaf; vm_page_t nkpg; while (kernel_vm_end <= addr) { if (nkpt == PAGE_SIZE/8 + PAGE_SIZE*PAGE_SIZE/64) panic("%s: out of kernel address space", __func__); dir1 = ia64_kptdir[KPTE_DIR0_INDEX(kernel_vm_end)]; if (dir1 == NULL) { nkpg = vm_page_alloc(NULL, nkpt++, VM_ALLOC_NOOBJ|VM_ALLOC_INTERRUPT|VM_ALLOC_WIRED); if (!nkpg) panic("%s: cannot add dir. page", __func__); dir1 = (struct ia64_lpte **) IA64_PHYS_TO_RR7(VM_PAGE_TO_PHYS(nkpg)); bzero(dir1, PAGE_SIZE); ia64_kptdir[KPTE_DIR0_INDEX(kernel_vm_end)] = dir1; } nkpg = vm_page_alloc(NULL, nkpt++, VM_ALLOC_NOOBJ|VM_ALLOC_INTERRUPT|VM_ALLOC_WIRED); if (!nkpg) panic("%s: cannot add PTE page", __func__); leaf = (struct ia64_lpte *) IA64_PHYS_TO_RR7(VM_PAGE_TO_PHYS(nkpg)); bzero(leaf, PAGE_SIZE); dir1[KPTE_DIR1_INDEX(kernel_vm_end)] = leaf; kernel_vm_end += PAGE_SIZE * NKPTEPG; } } /*************************************************** * page management routines. ***************************************************/ /* * free the pv_entry back to the free list */ static PMAP_INLINE void free_pv_entry(pv_entry_t pv) { pv_entry_count--; uma_zfree(pvzone, pv); } /* * get a new pv_entry, allocating a block from the system * when needed. */ static pv_entry_t get_pv_entry(pmap_t locked_pmap) { static const struct timeval printinterval = { 60, 0 }; static struct timeval lastprint; struct vpgqueues *vpq; struct ia64_lpte *pte; pmap_t oldpmap, pmap; pv_entry_t allocated_pv, next_pv, pv; vm_offset_t va; vm_page_t m; PMAP_LOCK_ASSERT(locked_pmap, MA_OWNED); mtx_assert(&vm_page_queue_mtx, MA_OWNED); allocated_pv = uma_zalloc(pvzone, M_NOWAIT); if (allocated_pv != NULL) { pv_entry_count++; if (pv_entry_count > pv_entry_high_water) pagedaemon_wakeup(); else return (allocated_pv); } /* * Reclaim pv entries: At first, destroy mappings to inactive * pages. After that, if a pv entry is still needed, destroy * mappings to active pages. */ if (ratecheck(&lastprint, &printinterval)) printf("Approaching the limit on PV entries, " "increase the vm.pmap.shpgperproc tunable.\n"); vpq = &vm_page_queues[PQ_INACTIVE]; retry: TAILQ_FOREACH(m, &vpq->pl, pageq) { if (m->hold_count || m->busy) continue; TAILQ_FOREACH_SAFE(pv, &m->md.pv_list, pv_list, next_pv) { va = pv->pv_va; pmap = pv->pv_pmap; /* Avoid deadlock and lock recursion. */ if (pmap > locked_pmap) PMAP_LOCK(pmap); else if (pmap != locked_pmap && !PMAP_TRYLOCK(pmap)) continue; pmap->pm_stats.resident_count--; oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(va); KASSERT(pte != NULL, ("pte")); pmap_remove_vhpt(va); pmap_invalidate_page(pmap, va); pmap_switch(oldpmap); if (pmap_accessed(pte)) vm_page_flag_set(m, PG_REFERENCED); if (pmap_dirty(pte)) vm_page_dirty(m); pmap_free_pte(pte, va); TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); m->md.pv_list_count--; TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); if (TAILQ_EMPTY(&m->md.pv_list)) vm_page_flag_clear(m, PG_WRITEABLE); if (pmap != locked_pmap) PMAP_UNLOCK(pmap); if (allocated_pv == NULL) allocated_pv = pv; else free_pv_entry(pv); } } if (allocated_pv == NULL) { if (vpq == &vm_page_queues[PQ_INACTIVE]) { vpq = &vm_page_queues[PQ_ACTIVE]; goto retry; } panic("get_pv_entry: increase the vm.pmap.shpgperproc tunable"); } return (allocated_pv); } /* * Conditionally create a pv entry. */ static boolean_t pmap_try_insert_pv_entry(pmap_t pmap, vm_offset_t va, vm_page_t m) { pv_entry_t pv; PMAP_LOCK_ASSERT(pmap, MA_OWNED); mtx_assert(&vm_page_queue_mtx, MA_OWNED); if (pv_entry_count < pv_entry_high_water && (pv = uma_zalloc(pvzone, M_NOWAIT)) != NULL) { pv_entry_count++; pv->pv_va = va; pv->pv_pmap = pmap; TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); m->md.pv_list_count++; return (TRUE); } else return (FALSE); } /* * Add an ia64_lpte to the VHPT. */ static void pmap_enter_vhpt(struct ia64_lpte *pte, vm_offset_t va) { struct ia64_bucket *bckt; struct ia64_lpte *vhpte; uint64_t pte_pa; /* Can fault, so get it out of the way. */ pte_pa = ia64_tpa((vm_offset_t)pte); vhpte = (struct ia64_lpte *)ia64_thash(va); bckt = (struct ia64_bucket *)vhpte->chain; mtx_lock_spin(&bckt->mutex); pte->chain = bckt->chain; ia64_mf(); bckt->chain = pte_pa; pmap_vhpt_inserts++; bckt->length++; mtx_unlock_spin(&bckt->mutex); } /* * Remove the ia64_lpte matching va from the VHPT. Return zero if it * worked or an appropriate error code otherwise. */ static int pmap_remove_vhpt(vm_offset_t va) { struct ia64_bucket *bckt; struct ia64_lpte *pte; struct ia64_lpte *lpte; struct ia64_lpte *vhpte; uint64_t chain, tag; tag = ia64_ttag(va); vhpte = (struct ia64_lpte *)ia64_thash(va); bckt = (struct ia64_bucket *)vhpte->chain; lpte = NULL; mtx_lock_spin(&bckt->mutex); chain = bckt->chain; pte = (struct ia64_lpte *)IA64_PHYS_TO_RR7(chain); while (chain != 0 && pte->tag != tag) { lpte = pte; chain = pte->chain; pte = (struct ia64_lpte *)IA64_PHYS_TO_RR7(chain); } if (chain == 0) { mtx_unlock_spin(&bckt->mutex); return (ENOENT); } /* Snip this pv_entry out of the collision chain. */ if (lpte == NULL) bckt->chain = pte->chain; else lpte->chain = pte->chain; ia64_mf(); bckt->length--; mtx_unlock_spin(&bckt->mutex); return (0); } /* * Find the ia64_lpte for the given va, if any. */ static struct ia64_lpte * pmap_find_vhpt(vm_offset_t va) { struct ia64_bucket *bckt; struct ia64_lpte *pte; uint64_t chain, tag; tag = ia64_ttag(va); pte = (struct ia64_lpte *)ia64_thash(va); bckt = (struct ia64_bucket *)pte->chain; mtx_lock_spin(&bckt->mutex); chain = bckt->chain; pte = (struct ia64_lpte *)IA64_PHYS_TO_RR7(chain); while (chain != 0 && pte->tag != tag) { chain = pte->chain; pte = (struct ia64_lpte *)IA64_PHYS_TO_RR7(chain); } mtx_unlock_spin(&bckt->mutex); return ((chain != 0) ? pte : NULL); } /* * Remove an entry from the list of managed mappings. */ static int pmap_remove_entry(pmap_t pmap, vm_page_t m, vm_offset_t va, pv_entry_t pv) { if (!pv) { if (m->md.pv_list_count < pmap->pm_stats.resident_count) { TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { if (pmap == pv->pv_pmap && va == pv->pv_va) break; } } else { TAILQ_FOREACH(pv, &pmap->pm_pvlist, pv_plist) { if (va == pv->pv_va) break; } } } if (pv) { TAILQ_REMOVE(&m->md.pv_list, pv, pv_list); m->md.pv_list_count--; if (TAILQ_FIRST(&m->md.pv_list) == NULL) vm_page_flag_clear(m, PG_WRITEABLE); TAILQ_REMOVE(&pmap->pm_pvlist, pv, pv_plist); free_pv_entry(pv); return 0; } else { return ENOENT; } } /* * Create a pv entry for page at pa for * (pmap, va). */ static void pmap_insert_entry(pmap_t pmap, vm_offset_t va, vm_page_t m) { pv_entry_t pv; pv = get_pv_entry(pmap); pv->pv_pmap = pmap; pv->pv_va = va; PMAP_LOCK_ASSERT(pmap, MA_OWNED); mtx_assert(&vm_page_queue_mtx, MA_OWNED); TAILQ_INSERT_TAIL(&pmap->pm_pvlist, pv, pv_plist); TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_list); m->md.pv_list_count++; } /* * Routine: pmap_extract * Function: * Extract the physical page address associated * with the given map/virtual_address pair. */ vm_paddr_t pmap_extract(pmap_t pmap, vm_offset_t va) { struct ia64_lpte *pte; pmap_t oldpmap; vm_paddr_t pa; pa = 0; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(va); if (pte != NULL && pmap_present(pte)) pa = pmap_ppn(pte); pmap_switch(oldpmap); PMAP_UNLOCK(pmap); return (pa); } /* * Routine: pmap_extract_and_hold * Function: * Atomically extract and hold the physical page * with the given pmap and virtual address pair * if that mapping permits the given protection. */ vm_page_t pmap_extract_and_hold(pmap_t pmap, vm_offset_t va, vm_prot_t prot) { struct ia64_lpte *pte; pmap_t oldpmap; vm_page_t m; m = NULL; vm_page_lock_queues(); PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(va); if (pte != NULL && pmap_present(pte) && (pmap_prot(pte) & prot) == prot) { m = PHYS_TO_VM_PAGE(pmap_ppn(pte)); vm_page_hold(m); } vm_page_unlock_queues(); pmap_switch(oldpmap); PMAP_UNLOCK(pmap); return (m); } /*************************************************** * Low level mapping routines..... ***************************************************/ /* * Find the kernel lpte for mapping the given virtual address, which * must be in the part of region 5 which we can cover with our kernel * 'page tables'. */ static struct ia64_lpte * pmap_find_kpte(vm_offset_t va) { struct ia64_lpte **dir1; struct ia64_lpte *leaf; KASSERT((va >> 61) == 5, ("kernel mapping 0x%lx not in region 5", va)); KASSERT(va < kernel_vm_end, ("kernel mapping 0x%lx out of range", va)); dir1 = ia64_kptdir[KPTE_DIR0_INDEX(va)]; leaf = dir1[KPTE_DIR1_INDEX(va)]; return (&leaf[KPTE_PTE_INDEX(va)]); } /* * Find a pte suitable for mapping a user-space address. If one exists * in the VHPT, that one will be returned, otherwise a new pte is * allocated. */ static struct ia64_lpte * pmap_find_pte(vm_offset_t va) { struct ia64_lpte *pte; if (va >= VM_MAXUSER_ADDRESS) return pmap_find_kpte(va); pte = pmap_find_vhpt(va); if (pte == NULL) { pte = uma_zalloc(ptezone, M_NOWAIT | M_ZERO); pte->tag = 1UL << 63; } return (pte); } /* * Free a pte which is now unused. This simply returns it to the zone * allocator if it is a user mapping. For kernel mappings, clear the * valid bit to make it clear that the mapping is not currently used. */ static void pmap_free_pte(struct ia64_lpte *pte, vm_offset_t va) { if (va < VM_MAXUSER_ADDRESS) uma_zfree(ptezone, pte); else pmap_clear_present(pte); } static PMAP_INLINE void pmap_pte_prot(pmap_t pm, struct ia64_lpte *pte, vm_prot_t prot) { static long prot2ar[4] = { PTE_AR_R, /* VM_PROT_NONE */ PTE_AR_RW, /* VM_PROT_WRITE */ PTE_AR_RX|PTE_ED, /* VM_PROT_EXECUTE */ PTE_AR_RWX|PTE_ED /* VM_PROT_WRITE|VM_PROT_EXECUTE */ }; pte->pte &= ~(PTE_PROT_MASK | PTE_PL_MASK | PTE_AR_MASK | PTE_ED); pte->pte |= (uint64_t)(prot & VM_PROT_ALL) << 56; pte->pte |= (prot == VM_PROT_NONE || pm == kernel_pmap) ? PTE_PL_KERN : PTE_PL_USER; pte->pte |= prot2ar[(prot & VM_PROT_ALL) >> 1]; } /* * Set a pte to contain a valid mapping and enter it in the VHPT. If * the pte was orginally valid, then its assumed to already be in the * VHPT. * This functions does not set the protection bits. It's expected * that those have been set correctly prior to calling this function. */ static void pmap_set_pte(struct ia64_lpte *pte, vm_offset_t va, vm_offset_t pa, boolean_t wired, boolean_t managed) { pte->pte &= PTE_PROT_MASK | PTE_PL_MASK | PTE_AR_MASK | PTE_ED; pte->pte |= PTE_PRESENT | PTE_MA_WB; pte->pte |= (managed) ? PTE_MANAGED : (PTE_DIRTY | PTE_ACCESSED); pte->pte |= (wired) ? PTE_WIRED : 0; pte->pte |= pa & PTE_PPN_MASK; pte->itir = PAGE_SHIFT << 2; pte->tag = ia64_ttag(va); } /* * Remove the (possibly managed) mapping represented by pte from the * given pmap. */ static int pmap_remove_pte(pmap_t pmap, struct ia64_lpte *pte, vm_offset_t va, pv_entry_t pv, int freepte) { int error; vm_page_t m; - KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(current_pmap)), + KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(md.current_pmap)), ("removing pte for non-current pmap")); /* * First remove from the VHPT. */ error = pmap_remove_vhpt(va); if (error) return (error); pmap_invalidate_page(pmap, va); if (pmap_wired(pte)) pmap->pm_stats.wired_count -= 1; pmap->pm_stats.resident_count -= 1; if (pmap_managed(pte)) { m = PHYS_TO_VM_PAGE(pmap_ppn(pte)); if (pmap_dirty(pte)) vm_page_dirty(m); if (pmap_accessed(pte)) vm_page_flag_set(m, PG_REFERENCED); error = pmap_remove_entry(pmap, m, va, pv); } if (freepte) pmap_free_pte(pte, va); return (error); } /* * Extract the physical page address associated with a kernel * virtual address. */ vm_paddr_t pmap_kextract(vm_offset_t va) { struct ia64_lpte *pte; vm_offset_t gwpage; KASSERT(va >= IA64_RR_BASE(5), ("Must be kernel VA")); /* Regions 6 and 7 are direct mapped. */ if (va >= IA64_RR_BASE(6)) return (IA64_RR_MASK(va)); /* EPC gateway page? */ gwpage = (vm_offset_t)ia64_get_k5(); if (va >= gwpage && va < gwpage + VM_GATEWAY_SIZE) return (IA64_RR_MASK((vm_offset_t)ia64_gateway_page)); /* Bail out if the virtual address is beyond our limits. */ if (va >= kernel_vm_end) return (0); pte = pmap_find_kpte(va); if (!pmap_present(pte)) return (0); return (pmap_ppn(pte) | (va & PAGE_MASK)); } /* * Add a list of wired pages to the kva this routine is only used for * temporary kernel mappings that do not need to have page modification * or references recorded. Note that old mappings are simply written * over. The page is effectively wired, but it's customary to not have * the PTE reflect that, nor update statistics. */ void pmap_qenter(vm_offset_t va, vm_page_t *m, int count) { struct ia64_lpte *pte; int i; for (i = 0; i < count; i++) { pte = pmap_find_kpte(va); if (pmap_present(pte)) pmap_invalidate_page(kernel_pmap, va); else pmap_enter_vhpt(pte, va); pmap_pte_prot(kernel_pmap, pte, VM_PROT_ALL); pmap_set_pte(pte, va, VM_PAGE_TO_PHYS(m[i]), FALSE, FALSE); va += PAGE_SIZE; } } /* * this routine jerks page mappings from the * kernel -- it is meant only for temporary mappings. */ void pmap_qremove(vm_offset_t va, int count) { struct ia64_lpte *pte; int i; for (i = 0; i < count; i++) { pte = pmap_find_kpte(va); if (pmap_present(pte)) { pmap_remove_vhpt(va); pmap_invalidate_page(kernel_pmap, va); pmap_clear_present(pte); } va += PAGE_SIZE; } } /* * Add a wired page to the kva. As for pmap_qenter(), it's customary * to not have the PTE reflect that, nor update statistics. */ void pmap_kenter(vm_offset_t va, vm_offset_t pa) { struct ia64_lpte *pte; pte = pmap_find_kpte(va); if (pmap_present(pte)) pmap_invalidate_page(kernel_pmap, va); else pmap_enter_vhpt(pte, va); pmap_pte_prot(kernel_pmap, pte, VM_PROT_ALL); pmap_set_pte(pte, va, pa, FALSE, FALSE); } /* * Remove a page from the kva */ void pmap_kremove(vm_offset_t va) { struct ia64_lpte *pte; pte = pmap_find_kpte(va); if (pmap_present(pte)) { pmap_remove_vhpt(va); pmap_invalidate_page(kernel_pmap, va); pmap_clear_present(pte); } } /* * Used to map a range of physical addresses into kernel * virtual address space. * * The value passed in '*virt' is a suggested virtual address for * the mapping. Architectures which can support a direct-mapped * physical to virtual region can return the appropriate address * within that region, leaving '*virt' unchanged. Other * architectures should map the pages starting at '*virt' and * update '*virt' with the first usable address after the mapped * region. */ vm_offset_t pmap_map(vm_offset_t *virt, vm_offset_t start, vm_offset_t end, int prot) { return IA64_PHYS_TO_RR7(start); } /* * Remove a single page from a process address space */ static void pmap_remove_page(pmap_t pmap, vm_offset_t va) { struct ia64_lpte *pte; - KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(current_pmap)), + KASSERT((pmap == kernel_pmap || pmap == PCPU_GET(md.current_pmap)), ("removing page for non-current pmap")); pte = pmap_find_vhpt(va); if (pte != NULL) pmap_remove_pte(pmap, pte, va, 0, 1); return; } /* * Remove the given range of addresses from the specified map. * * It is assumed that the start and end are properly * rounded to the page size. */ void pmap_remove(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { pmap_t oldpmap; vm_offset_t va; pv_entry_t npv, pv; struct ia64_lpte *pte; if (pmap->pm_stats.resident_count == 0) return; vm_page_lock_queues(); PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); /* * special handling of removing one page. a very * common operation and easy to short circuit some * code. */ if (sva + PAGE_SIZE == eva) { pmap_remove_page(pmap, sva); goto out; } if (pmap->pm_stats.resident_count < ((eva - sva) >> PAGE_SHIFT)) { TAILQ_FOREACH_SAFE(pv, &pmap->pm_pvlist, pv_plist, npv) { va = pv->pv_va; if (va >= sva && va < eva) { pte = pmap_find_vhpt(va); KASSERT(pte != NULL, ("pte")); pmap_remove_pte(pmap, pte, va, pv, 1); } } } else { for (va = sva; va < eva; va += PAGE_SIZE) { pte = pmap_find_vhpt(va); if (pte != NULL) pmap_remove_pte(pmap, pte, va, 0, 1); } } out: vm_page_unlock_queues(); pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } /* * Routine: pmap_remove_all * Function: * Removes this physical page from * all physical maps in which it resides. * Reflects back modify bits to the pager. * * Notes: * Original versions of this routine were very * inefficient because they iteratively called * pmap_remove (slow...) */ void pmap_remove_all(vm_page_t m) { pmap_t oldpmap; pv_entry_t pv; #if defined(DIAGNOSTIC) /* * XXX This makes pmap_remove_all() illegal for non-managed pages! */ if (m->flags & PG_FICTITIOUS) { panic("pmap_remove_all: illegal for unmanaged page, va: 0x%lx", VM_PAGE_TO_PHYS(m)); } #endif mtx_assert(&vm_page_queue_mtx, MA_OWNED); while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { struct ia64_lpte *pte; pmap_t pmap = pv->pv_pmap; vm_offset_t va = pv->pv_va; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(va); KASSERT(pte != NULL, ("pte")); if (pmap_ppn(pte) != VM_PAGE_TO_PHYS(m)) panic("pmap_remove_all: pv_table for %lx is inconsistent", VM_PAGE_TO_PHYS(m)); pmap_remove_pte(pmap, pte, va, pv, 1); pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } vm_page_flag_clear(m, PG_WRITEABLE); } /* * Set the physical protection on the * specified range of this map as requested. */ void pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) { pmap_t oldpmap; struct ia64_lpte *pte; if ((prot & VM_PROT_READ) == VM_PROT_NONE) { pmap_remove(pmap, sva, eva); return; } if ((prot & (VM_PROT_WRITE|VM_PROT_EXECUTE)) == (VM_PROT_WRITE|VM_PROT_EXECUTE)) return; if ((sva & PAGE_MASK) || (eva & PAGE_MASK)) panic("pmap_protect: unaligned addresses"); vm_page_lock_queues(); PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); for ( ; sva < eva; sva += PAGE_SIZE) { /* If page is invalid, skip this page */ pte = pmap_find_vhpt(sva); if (pte == NULL) continue; /* If there's no change, skip it too */ if (pmap_prot(pte) == prot) continue; if (pmap_managed(pte)) { vm_offset_t pa = pmap_ppn(pte); vm_page_t m = PHYS_TO_VM_PAGE(pa); if (pmap_dirty(pte)) { vm_page_dirty(m); pmap_clear_dirty(pte); } if (pmap_accessed(pte)) { vm_page_flag_set(m, PG_REFERENCED); pmap_clear_accessed(pte); } } if (prot & VM_PROT_EXECUTE) ia64_sync_icache(sva, PAGE_SIZE); pmap_pte_prot(pmap, pte, prot); pmap_invalidate_page(pmap, sva); } vm_page_unlock_queues(); pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } /* * Insert the given physical page (p) at * the specified virtual address (v) in the * target physical map with the protection requested. * * If specified, the page will be wired down, meaning * that the related pte can not be reclaimed. * * NB: This is the only routine which MAY NOT lazy-evaluate * or lose information. That is, this routine must actually * insert this page into the given map NOW. */ void pmap_enter(pmap_t pmap, vm_offset_t va, vm_prot_t access, vm_page_t m, vm_prot_t prot, boolean_t wired) { pmap_t oldpmap; vm_offset_t pa; vm_offset_t opa; struct ia64_lpte origpte; struct ia64_lpte *pte; boolean_t icache_inval, managed; vm_page_lock_queues(); PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); va &= ~PAGE_MASK; #ifdef DIAGNOSTIC if (va > VM_MAX_KERNEL_ADDRESS) panic("pmap_enter: toobig"); #endif /* * Find (or create) a pte for the given mapping. */ while ((pte = pmap_find_pte(va)) == NULL) { pmap_switch(oldpmap); PMAP_UNLOCK(pmap); vm_page_unlock_queues(); VM_WAIT; vm_page_lock_queues(); PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); } origpte = *pte; if (!pmap_present(pte)) { opa = ~0UL; pmap_enter_vhpt(pte, va); } else opa = pmap_ppn(pte); managed = FALSE; pa = VM_PAGE_TO_PHYS(m); icache_inval = (prot & VM_PROT_EXECUTE) ? TRUE : FALSE; /* * Mapping has not changed, must be protection or wiring change. */ if (opa == pa) { /* * Wiring change, just update stats. We don't worry about * wiring PT pages as they remain resident as long as there * are valid mappings in them. Hence, if a user page is wired, * the PT page will be also. */ if (wired && !pmap_wired(&origpte)) pmap->pm_stats.wired_count++; else if (!wired && pmap_wired(&origpte)) pmap->pm_stats.wired_count--; managed = (pmap_managed(&origpte)) ? TRUE : FALSE; /* * We might be turning off write access to the page, * so we go ahead and sense modify status. Otherwise, * we can avoid I-cache invalidation if the page * already allowed execution. */ if (managed && pmap_dirty(&origpte)) vm_page_dirty(m); else if (pmap_exec(&origpte)) icache_inval = FALSE; pmap_invalidate_page(pmap, va); goto validate; } /* * Mapping has changed, invalidate old range and fall * through to handle validating new mapping. */ if (opa != ~0UL) { pmap_remove_pte(pmap, pte, va, 0, 0); pmap_enter_vhpt(pte, va); } /* * Enter on the PV list if part of our managed memory. */ if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0) { KASSERT(va < kmi.clean_sva || va >= kmi.clean_eva, ("pmap_enter: managed mapping within the clean submap")); pmap_insert_entry(pmap, va, m); managed = TRUE; } /* * Increment counters */ pmap->pm_stats.resident_count++; if (wired) pmap->pm_stats.wired_count++; validate: /* * Now validate mapping with desired protection/wiring. This * adds the pte to the VHPT if necessary. */ pmap_pte_prot(pmap, pte, prot); pmap_set_pte(pte, va, pa, wired, managed); /* Invalidate the I-cache when needed. */ if (icache_inval) ia64_sync_icache(va, PAGE_SIZE); if ((prot & VM_PROT_WRITE) != 0) vm_page_flag_set(m, PG_WRITEABLE); vm_page_unlock_queues(); pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } /* * Maps a sequence of resident pages belonging to the same object. * The sequence begins with the given page m_start. This page is * mapped at the given virtual address start. Each subsequent page is * mapped at a virtual address that is offset from start by the same * amount as the page is offset from m_start within the object. The * last page in the sequence is the page with the largest offset from * m_start that can be mapped at a virtual address less than the given * virtual address end. Not every virtual page between start and end * is mapped; only those for which a resident page exists with the * corresponding offset from m_start are mapped. */ void pmap_enter_object(pmap_t pmap, vm_offset_t start, vm_offset_t end, vm_page_t m_start, vm_prot_t prot) { pmap_t oldpmap; vm_page_t m; vm_pindex_t diff, psize; VM_OBJECT_LOCK_ASSERT(m_start->object, MA_OWNED); psize = atop(end - start); m = m_start; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) { pmap_enter_quick_locked(pmap, start + ptoa(diff), m, prot); m = TAILQ_NEXT(m, listq); } pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } /* * this code makes some *MAJOR* assumptions: * 1. Current pmap & pmap exists. * 2. Not wired. * 3. Read access. * 4. No page table pages. * but is *MUCH* faster than pmap_enter... */ void pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot) { pmap_t oldpmap; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pmap_enter_quick_locked(pmap, va, m, prot); pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } static void pmap_enter_quick_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot) { struct ia64_lpte *pte; boolean_t managed; KASSERT(va < kmi.clean_sva || va >= kmi.clean_eva || (m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) != 0, ("pmap_enter_quick_locked: managed mapping within the clean submap")); mtx_assert(&vm_page_queue_mtx, MA_OWNED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); if ((pte = pmap_find_pte(va)) == NULL) return; if (!pmap_present(pte)) { /* Enter on the PV list if the page is managed. */ if ((m->flags & (PG_FICTITIOUS | PG_UNMANAGED)) == 0) { if (!pmap_try_insert_pv_entry(pmap, va, m)) { pmap_free_pte(pte, va); return; } managed = TRUE; } else managed = FALSE; /* Increment counters. */ pmap->pm_stats.resident_count++; /* Initialise with R/O protection and enter into VHPT. */ pmap_enter_vhpt(pte, va); pmap_pte_prot(pmap, pte, prot & (VM_PROT_READ | VM_PROT_EXECUTE)); pmap_set_pte(pte, va, VM_PAGE_TO_PHYS(m), FALSE, managed); if (prot & VM_PROT_EXECUTE) ia64_sync_icache(va, PAGE_SIZE); } } /* * pmap_object_init_pt preloads the ptes for a given object * into the specified pmap. This eliminates the blast of soft * faults on process startup and immediately after an mmap. */ void pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_object_t object, vm_pindex_t pindex, vm_size_t size) { VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG, ("pmap_object_init_pt: non-device object")); } /* * Routine: pmap_change_wiring * Function: Change the wiring attribute for a map/virtual-address * pair. * In/out conditions: * The mapping must already exist in the pmap. */ void pmap_change_wiring(pmap, va, wired) register pmap_t pmap; vm_offset_t va; boolean_t wired; { pmap_t oldpmap; struct ia64_lpte *pte; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(va); KASSERT(pte != NULL, ("pte")); if (wired && !pmap_wired(pte)) { pmap->pm_stats.wired_count++; pmap_set_wired(pte); } else if (!wired && pmap_wired(pte)) { pmap->pm_stats.wired_count--; pmap_clear_wired(pte); } pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } /* * Copy the range specified by src_addr/len * from the source map to the range dst_addr/len * in the destination map. * * This routine is only advisory and need not do anything. */ void pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr) { } /* * pmap_zero_page zeros the specified hardware page by * mapping it into virtual memory and using bzero to clear * its contents. */ void pmap_zero_page(vm_page_t m) { vm_offset_t va = IA64_PHYS_TO_RR7(VM_PAGE_TO_PHYS(m)); bzero((caddr_t) va, PAGE_SIZE); } /* * pmap_zero_page_area zeros the specified hardware page by * mapping it into virtual memory and using bzero to clear * its contents. * * off and size must reside within a single page. */ void pmap_zero_page_area(vm_page_t m, int off, int size) { vm_offset_t va = IA64_PHYS_TO_RR7(VM_PAGE_TO_PHYS(m)); bzero((char *)(caddr_t)va + off, size); } /* * pmap_zero_page_idle zeros the specified hardware page by * mapping it into virtual memory and using bzero to clear * its contents. This is for the vm_idlezero process. */ void pmap_zero_page_idle(vm_page_t m) { vm_offset_t va = IA64_PHYS_TO_RR7(VM_PAGE_TO_PHYS(m)); bzero((caddr_t) va, PAGE_SIZE); } /* * pmap_copy_page copies the specified (machine independent) * page by mapping the page into virtual memory and using * bcopy to copy the page, one machine dependent page at a * time. */ void pmap_copy_page(vm_page_t msrc, vm_page_t mdst) { vm_offset_t src = IA64_PHYS_TO_RR7(VM_PAGE_TO_PHYS(msrc)); vm_offset_t dst = IA64_PHYS_TO_RR7(VM_PAGE_TO_PHYS(mdst)); bcopy((caddr_t) src, (caddr_t) dst, PAGE_SIZE); } /* * Returns true if the pmap's pv is one of the first * 16 pvs linked to from this page. This count may * be changed upwards or downwards in the future; it * is only necessary that true be returned for a small * subset of pmaps for proper page aging. */ boolean_t pmap_page_exists_quick(pmap_t pmap, vm_page_t m) { pv_entry_t pv; int loops = 0; if (m->flags & PG_FICTITIOUS) return FALSE; /* * Not found, check current mappings returning immediately if found. */ mtx_assert(&vm_page_queue_mtx, MA_OWNED); TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { if (pv->pv_pmap == pmap) { return TRUE; } loops++; if (loops >= 16) break; } return (FALSE); } /* * pmap_page_wired_mappings: * * Return the number of managed mappings to the given physical page * that are wired. */ int pmap_page_wired_mappings(vm_page_t m) { struct ia64_lpte *pte; pmap_t oldpmap, pmap; pv_entry_t pv; int count; count = 0; if ((m->flags & PG_FICTITIOUS) != 0) return (count); mtx_assert(&vm_page_queue_mtx, MA_OWNED); TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { pmap = pv->pv_pmap; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(pv->pv_va); KASSERT(pte != NULL, ("pte")); if (pmap_wired(pte)) count++; pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } return (count); } /* * Remove all pages from specified address space * this aids process exit speeds. Also, this code * is special cased for current process only, but * can have the more generic (and slightly slower) * mode enabled. This is much faster than pmap_remove * in the case of running down an entire address space. */ void pmap_remove_pages(pmap_t pmap) { pmap_t oldpmap; pv_entry_t pv, npv; if (pmap != vmspace_pmap(curthread->td_proc->p_vmspace)) { printf("warning: pmap_remove_pages called with non-current pmap\n"); return; } vm_page_lock_queues(); PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); for (pv = TAILQ_FIRST(&pmap->pm_pvlist); pv; pv = npv) { struct ia64_lpte *pte; npv = TAILQ_NEXT(pv, pv_plist); pte = pmap_find_vhpt(pv->pv_va); KASSERT(pte != NULL, ("pte")); if (!pmap_wired(pte)) pmap_remove_pte(pmap, pte, pv->pv_va, pv, 1); } pmap_switch(oldpmap); PMAP_UNLOCK(pmap); vm_page_unlock_queues(); } /* * pmap_ts_referenced: * * Return a count of reference bits for a page, clearing those bits. * It is not necessary for every reference bit to be cleared, but it * is necessary that 0 only be returned when there are truly no * reference bits set. * * XXX: The exact number of bits to check and clear is a matter that * should be tested and standardized at some point in the future for * optimal aging of shared pages. */ int pmap_ts_referenced(vm_page_t m) { struct ia64_lpte *pte; pmap_t oldpmap; pv_entry_t pv; int count = 0; if (m->flags & PG_FICTITIOUS) return 0; TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { PMAP_LOCK(pv->pv_pmap); oldpmap = pmap_switch(pv->pv_pmap); pte = pmap_find_vhpt(pv->pv_va); KASSERT(pte != NULL, ("pte")); if (pmap_accessed(pte)) { count++; pmap_clear_accessed(pte); pmap_invalidate_page(pv->pv_pmap, pv->pv_va); } pmap_switch(oldpmap); PMAP_UNLOCK(pv->pv_pmap); } return count; } /* * pmap_is_modified: * * Return whether or not the specified physical page was modified * in any physical maps. */ boolean_t pmap_is_modified(vm_page_t m) { struct ia64_lpte *pte; pmap_t oldpmap; pv_entry_t pv; boolean_t rv; rv = FALSE; if (m->flags & PG_FICTITIOUS) return (rv); TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { PMAP_LOCK(pv->pv_pmap); oldpmap = pmap_switch(pv->pv_pmap); pte = pmap_find_vhpt(pv->pv_va); pmap_switch(oldpmap); KASSERT(pte != NULL, ("pte")); rv = pmap_dirty(pte) ? TRUE : FALSE; PMAP_UNLOCK(pv->pv_pmap); if (rv) break; } return (rv); } /* * pmap_is_prefaultable: * * Return whether or not the specified virtual address is elgible * for prefault. */ boolean_t pmap_is_prefaultable(pmap_t pmap, vm_offset_t addr) { struct ia64_lpte *pte; pte = pmap_find_vhpt(addr); if (pte != NULL && pmap_present(pte)) return (FALSE); return (TRUE); } /* * Clear the modify bits on the specified physical page. */ void pmap_clear_modify(vm_page_t m) { struct ia64_lpte *pte; pmap_t oldpmap; pv_entry_t pv; if (m->flags & PG_FICTITIOUS) return; TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { PMAP_LOCK(pv->pv_pmap); oldpmap = pmap_switch(pv->pv_pmap); pte = pmap_find_vhpt(pv->pv_va); KASSERT(pte != NULL, ("pte")); if (pmap_dirty(pte)) { pmap_clear_dirty(pte); pmap_invalidate_page(pv->pv_pmap, pv->pv_va); } pmap_switch(oldpmap); PMAP_UNLOCK(pv->pv_pmap); } } /* * pmap_clear_reference: * * Clear the reference bit on the specified physical page. */ void pmap_clear_reference(vm_page_t m) { struct ia64_lpte *pte; pmap_t oldpmap; pv_entry_t pv; if (m->flags & PG_FICTITIOUS) return; TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { PMAP_LOCK(pv->pv_pmap); oldpmap = pmap_switch(pv->pv_pmap); pte = pmap_find_vhpt(pv->pv_va); KASSERT(pte != NULL, ("pte")); if (pmap_accessed(pte)) { pmap_clear_accessed(pte); pmap_invalidate_page(pv->pv_pmap, pv->pv_va); } pmap_switch(oldpmap); PMAP_UNLOCK(pv->pv_pmap); } } /* * Clear the write and modified bits in each of the given page's mappings. */ void pmap_remove_write(vm_page_t m) { struct ia64_lpte *pte; pmap_t oldpmap, pmap; pv_entry_t pv; vm_prot_t prot; mtx_assert(&vm_page_queue_mtx, MA_OWNED); if ((m->flags & PG_FICTITIOUS) != 0 || (m->flags & PG_WRITEABLE) == 0) return; TAILQ_FOREACH(pv, &m->md.pv_list, pv_list) { pmap = pv->pv_pmap; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(pv->pv_va); KASSERT(pte != NULL, ("pte")); prot = pmap_prot(pte); if ((prot & VM_PROT_WRITE) != 0) { if (pmap_dirty(pte)) { vm_page_dirty(m); pmap_clear_dirty(pte); } prot &= ~VM_PROT_WRITE; pmap_pte_prot(pmap, pte, prot); pmap_invalidate_page(pmap, pv->pv_va); } pmap_switch(oldpmap); PMAP_UNLOCK(pmap); } vm_page_flag_clear(m, PG_WRITEABLE); } /* * Map a set of physical memory pages into the kernel virtual * address space. Return a pointer to where it is mapped. This * routine is intended to be used for mapping device memory, * NOT real memory. */ void * pmap_mapdev(vm_offset_t pa, vm_size_t size) { return (void*) IA64_PHYS_TO_RR6(pa); } /* * 'Unmap' a range mapped by pmap_mapdev(). */ void pmap_unmapdev(vm_offset_t va, vm_size_t size) { return; } /* * perform the pmap work for mincore */ int pmap_mincore(pmap_t pmap, vm_offset_t addr) { pmap_t oldpmap; struct ia64_lpte *pte, tpte; int val = 0; PMAP_LOCK(pmap); oldpmap = pmap_switch(pmap); pte = pmap_find_vhpt(addr); if (pte != NULL) { tpte = *pte; pte = &tpte; } pmap_switch(oldpmap); PMAP_UNLOCK(pmap); if (pte == NULL) return 0; if (pmap_present(pte)) { vm_page_t m; vm_offset_t pa; val = MINCORE_INCORE; if (!pmap_managed(pte)) return val; pa = pmap_ppn(pte); m = PHYS_TO_VM_PAGE(pa); /* * Modified by us */ if (pmap_dirty(pte)) val |= MINCORE_MODIFIED|MINCORE_MODIFIED_OTHER; else { /* * Modified by someone */ vm_page_lock_queues(); if (pmap_is_modified(m)) val |= MINCORE_MODIFIED_OTHER; vm_page_unlock_queues(); } /* * Referenced by us */ if (pmap_accessed(pte)) val |= MINCORE_REFERENCED|MINCORE_REFERENCED_OTHER; else { /* * Referenced by someone */ vm_page_lock_queues(); if (pmap_ts_referenced(m)) { val |= MINCORE_REFERENCED_OTHER; vm_page_flag_set(m, PG_REFERENCED); } vm_page_unlock_queues(); } } return val; } void pmap_activate(struct thread *td) { pmap_switch(vmspace_pmap(td->td_proc->p_vmspace)); } pmap_t pmap_switch(pmap_t pm) { pmap_t prevpm; int i; critical_enter(); - prevpm = PCPU_GET(current_pmap); + prevpm = PCPU_GET(md.current_pmap); if (prevpm == pm) goto out; if (prevpm != NULL) atomic_clear_32(&prevpm->pm_active, PCPU_GET(cpumask)); if (pm == NULL) { for (i = 0; i < 5; i++) { ia64_set_rr(IA64_RR_BASE(i), (i << 8)|(PAGE_SHIFT << 2)|1); } } else { for (i = 0; i < 5; i++) { ia64_set_rr(IA64_RR_BASE(i), (pm->pm_rid[i] << 8)|(PAGE_SHIFT << 2)|1); } atomic_set_32(&pm->pm_active, PCPU_GET(cpumask)); } - PCPU_SET(current_pmap, pm); + PCPU_SET(md.current_pmap, pm); ia64_srlz_d(); out: critical_exit(); return (prevpm); } void pmap_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz) { pmap_t oldpm; struct ia64_lpte *pte; vm_offset_t lim; vm_size_t len; sz += va & 31; va &= ~31; sz = (sz + 31) & ~31; PMAP_LOCK(pm); oldpm = pmap_switch(pm); while (sz > 0) { lim = round_page(va); len = MIN(lim - va, sz); pte = pmap_find_vhpt(va); if (pte != NULL && pmap_present(pte)) ia64_sync_icache(va, len); va += len; sz -= len; } pmap_switch(oldpm); PMAP_UNLOCK(pm); } /* * Increase the starting virtual address of the given mapping if a * different alignment might result in more superpage mappings. */ void pmap_align_superpage(vm_object_t object, vm_ooffset_t offset, vm_offset_t *addr, vm_size_t size) { } #include "opt_ddb.h" #ifdef DDB #include static const char* psnames[] = { "1B", "2B", "4B", "8B", "16B", "32B", "64B", "128B", "256B", "512B", "1K", "2K", "4K", "8K", "16K", "32K", "64K", "128K", "256K", "512K", "1M", "2M", "4M", "8M", "16M", "32M", "64M", "128M", "256M", "512M", "1G", "2G" }; static void print_trs(int type) { struct ia64_pal_result res; int i, maxtr; struct { pt_entry_t pte; uint64_t itir; uint64_t ifa; struct ia64_rr rr; } buf; static const char *manames[] = { "WB", "bad", "bad", "bad", "UC", "UCE", "WC", "NaT", }; res = ia64_call_pal_static(PAL_VM_SUMMARY, 0, 0, 0); if (res.pal_status != 0) { db_printf("Can't get VM summary\n"); return; } if (type == 0) maxtr = (res.pal_result[0] >> 40) & 0xff; else maxtr = (res.pal_result[0] >> 32) & 0xff; db_printf("V RID Virtual Page Physical Page PgSz ED AR PL D A MA P KEY\n"); for (i = 0; i <= maxtr; i++) { bzero(&buf, sizeof(buf)); res = ia64_call_pal_stacked_physical (PAL_VM_TR_READ, i, type, ia64_tpa((uint64_t) &buf)); if (!(res.pal_result[0] & 1)) buf.pte &= ~PTE_AR_MASK; if (!(res.pal_result[0] & 2)) buf.pte &= ~PTE_PL_MASK; if (!(res.pal_result[0] & 4)) pmap_clear_dirty(&buf); if (!(res.pal_result[0] & 8)) buf.pte &= ~PTE_MA_MASK; db_printf("%d %06x %013lx %013lx %4s %d %d %d %d %d %-3s " "%d %06x\n", (int)buf.ifa & 1, buf.rr.rr_rid, buf.ifa >> 12, (buf.pte & PTE_PPN_MASK) >> 12, psnames[(buf.itir & ITIR_PS_MASK) >> 2], (buf.pte & PTE_ED) ? 1 : 0, (int)(buf.pte & PTE_AR_MASK) >> 9, (int)(buf.pte & PTE_PL_MASK) >> 7, (pmap_dirty(&buf)) ? 1 : 0, (pmap_accessed(&buf)) ? 1 : 0, manames[(buf.pte & PTE_MA_MASK) >> 2], (pmap_present(&buf)) ? 1 : 0, (int)((buf.itir & ITIR_KEY_MASK) >> 8)); } } DB_COMMAND(itr, db_itr) { print_trs(0); } DB_COMMAND(dtr, db_dtr) { print_trs(1); } DB_COMMAND(rr, db_rr) { int i; uint64_t t; struct ia64_rr rr; printf("RR RID PgSz VE\n"); for (i = 0; i < 8; i++) { __asm __volatile ("mov %0=rr[%1]" : "=r"(t) : "r"(IA64_RR_BASE(i))); *(uint64_t *) &rr = t; printf("%d %06x %4s %d\n", i, rr.rr_rid, psnames[rr.rr_ps], rr.rr_ve); } } DB_COMMAND(thash, db_thash) { if (!have_addr) return; db_printf("%p\n", (void *) ia64_thash(addr)); } DB_COMMAND(ttag, db_ttag) { if (!have_addr) return; db_printf("0x%lx\n", ia64_ttag(addr)); } DB_COMMAND(kpte, db_kpte) { struct ia64_lpte *pte; if (!have_addr) { db_printf("usage: kpte \n"); return; } if (addr < VM_MIN_KERNEL_ADDRESS) { db_printf("kpte: error: invalid \n"); return; } pte = pmap_find_kpte(addr); db_printf("kpte at %p:\n", pte); db_printf(" pte =%016lx\n", pte->pte); db_printf(" itir =%016lx\n", pte->itir); db_printf(" tag =%016lx\n", pte->tag); db_printf(" chain=%016lx\n", pte->chain); } #endif diff --git a/sys/ia64/include/kdb.h b/sys/ia64/include/kdb.h index d8a12e037b4c..8a9cc3a1e3b3 100644 --- a/sys/ia64/include/kdb.h +++ b/sys/ia64/include/kdb.h @@ -1,75 +1,75 @@ /*- * Copyright (c) 2004 Marcel Moolenaar * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_KDB_H_ #define _MACHINE_KDB_H_ #include #include #include -#define KDB_STOPPEDPCB(pc) (&(pc)->pc_pcb) +#define KDB_STOPPEDPCB(pc) (&(pc)->pc_md.pcb) static __inline void kdb_cpu_clear_singlestep(void) { kdb_frame->tf_special.psr &= ~IA64_PSR_SS; } static __inline void kdb_cpu_set_singlestep(void) { kdb_frame->tf_special.psr |= IA64_PSR_SS; } static __inline void kdb_cpu_sync_icache(unsigned char *addr, size_t size) { vm_offset_t cacheline; cacheline = (uintptr_t)addr & ~31; size += (uintptr_t)addr - cacheline; size = (size + 31) & ~31; while (size > 0) { __asm __volatile("fc %0;; sync.i;; srlz.i;;" :: "r"(cacheline)); cacheline += 32; size -= 32; } } static __inline void kdb_cpu_trap(int vector, int _) { __asm __volatile("flushrs;;"); if (vector == IA64_VEC_BREAK && kdb_frame->tf_special.ifa == IA64_FIXED_BREAK) kdb_frame->tf_special.psr += IA64_PSR_RI_1; } #endif /* _MACHINE_KDB_H_ */ diff --git a/sys/ia64/include/pcpu.h b/sys/ia64/include/pcpu.h index c6c999b30154..bc9fe0813d7c 100644 --- a/sys/ia64/include/pcpu.h +++ b/sys/ia64/include/pcpu.h @@ -1,85 +1,88 @@ /*- * Copyright (c) 1999 Luoqi Chen * Copyright (c) Peter Wemm * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _MACHINE_PCPU_H_ #define _MACHINE_PCPU_H_ #include #include struct pcpu_stats { u_long pcs_nasts; /* IPI_AST counter. */ u_long pcs_nclks; /* Clock interrupt counter. */ u_long pcs_nextints; /* ExtINT counter. */ u_long pcs_nhighfps; /* IPI_HIGH_FP counter. */ u_long pcs_nhwints; /* Hardware int. counter. */ u_long pcs_npreempts; /* IPI_PREEMPT counter. */ u_long pcs_nrdvs; /* IPI_RENDEZVOUS counter. */ u_long pcs_nstops; /* IPI_STOP counter. */ u_long pcs_nstrays; /* Stray interrupt counter. */ +}; +struct pcpu_md { + struct pcb pcb; /* Used by IPI_STOP */ + struct pmap *current_pmap; /* active pmap */ + vm_offset_t vhpt; /* Address of VHPT */ + uint64_t lid; /* local CPU ID */ + uint64_t clock; /* Clock counter. */ + uint64_t clockadj; /* Clock adjust. */ + uint32_t awake:1; /* CPU is awake? */ + struct pcpu_stats stats; /* Interrupt stats. */ #ifdef _KERNEL - struct sysctl_ctx_list pcs_sysctl_ctx; - struct sysctl_oid *pcs_sysctl_tree; + struct sysctl_ctx_list sysctl_ctx; + struct sysctl_oid *sysctl_tree; #endif }; #define PCPU_MD_FIELDS \ - struct pcb pc_pcb; /* Used by IPI_STOP */ \ - struct pmap *pc_current_pmap; /* active pmap */ \ - vm_offset_t pc_vhpt; /* Address of VHPT */ \ - uint64_t pc_lid; /* local CPU ID */ \ - uint64_t pc_clock; /* Clock counter. */ \ - uint64_t pc_clockadj; /* Clock adjust. */ \ - uint32_t pc_awake:1; /* CPU is awake? */ \ uint32_t pc_acpi_id; /* ACPI CPU id. */ \ - struct pcpu_stats pc_stats + struct pcpu_md pc_md /* MD fields. */ #ifdef _KERNEL struct pcpu; register struct pcpu *pcpup __asm__("r13"); #define PCPU_GET(member) (pcpup->pc_ ## member) /* * XXX The implementation of this operation should be made atomic * with respect to preemption. */ #define PCPU_ADD(member, value) (pcpup->pc_ ## member += (value)) #define PCPU_INC(member) PCPU_ADD(member, 1) #define PCPU_PTR(member) (&pcpup->pc_ ## member) #define PCPU_SET(member,value) (pcpup->pc_ ## member = (value)) void pcpu_initclock(void); #endif /* _KERNEL */ #endif /* !_MACHINE_PCPU_H_ */