/* * Initialize the local APIC on the BSP. */ static int mptable_setup_local(void) { PCPU_SET(apic_id, 0); PCPU_SET(vcpu_id, 0); return (0); }
void pcpu_initclock(void) { PCPU_SET(clockadj, 0); PCPU_SET(clock, ia64_get_itc()); ia64_set_itm(PCPU_GET(clock) + ia64_clock_reload); ia64_set_itv(CLOCK_VECTOR); /* highest priority class */ ia64_srlz_d(); }
/* * Map the local APIC and setup necessary interrupt vectors. */ void lapic_init(vm_paddr_t addr) { u_int regs[4]; int i, arat; /* Map the local APIC and setup the spurious interrupt handler. */ KASSERT(trunc_page(addr) == addr, ("local APIC not aligned on a page boundary")); lapic_paddr = addr; lapic = pmap_mapdev(addr, sizeof(lapic_t)); setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_APIC, SEL_KPL, GSEL_APIC); /* Perform basic initialization of the BSP's local APIC. */ lapic_enable(); /* Set BSP's per-CPU local APIC ID. */ PCPU_SET(apic_id, lapic_id()); /* Local APIC timer interrupt. */ setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_APIC, SEL_KPL, GSEL_APIC); /* Local APIC error interrupt. */ setidt(APIC_ERROR_INT, IDTVEC(errorint), SDT_APIC, SEL_KPL, GSEL_APIC); /* XXX: Thermal interrupt */ /* Local APIC CMCI. */ setidt(APIC_CMC_INT, IDTVEC(cmcint), SDT_APICT, SEL_KPL, GSEL_APIC); if ((resource_int_value("apic", 0, "clock", &i) != 0 || i != 0)) { arat = 0; /* Intel CPUID 0x06 EAX[2] set if APIC timer runs in C3. */ if (cpu_vendor_id == CPU_VENDOR_INTEL && cpu_high >= 6) { do_cpuid(0x06, regs); if ((regs[0] & CPUTPM1_ARAT) != 0) arat = 1; } bzero(&lapic_et, sizeof(lapic_et)); lapic_et.et_name = "LAPIC"; lapic_et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT | ET_FLAGS_PERCPU; lapic_et.et_quality = 600; if (!arat) { lapic_et.et_flags |= ET_FLAGS_C3STOP; lapic_et.et_quality -= 200; } lapic_et.et_frequency = 0; /* We don't know frequency yet, so trying to guess. */ lapic_et.et_min_period.sec = 0; lapic_et.et_min_period.frac = 0x00001000LL << 32; lapic_et.et_max_period.sec = 1; lapic_et.et_max_period.frac = 0; lapic_et.et_start = lapic_et_start; lapic_et.et_stop = lapic_et_stop; lapic_et.et_priv = NULL; et_register(&lapic_et); } }
/* * Create an environment for the EFI runtime code call. The most * important part is creating the required 1:1 physical->virtual * mappings for the runtime segments. To do that, we manually create * page table which unmap userspace but gives correct kernel mapping. * The 1:1 mappings for runtime segments usually occupy low 4G of the * physical address map. * * The 1:1 mappings were chosen over the SetVirtualAddressMap() EFI RT * service, because there are some BIOSes which fail to correctly * relocate itself on the call, requiring both 1:1 and virtual * mapping. As result, we must provide 1:1 mapping anyway, so no * reason to bother with the virtual map, and no need to add a * complexity into loader. * * The fpu_kern_enter() call allows firmware to use FPU, as mandated * by the specification. In particular, CR0.TS bit is cleared. Also * it enters critical section, giving us neccessary protection against * context switch. * * There is no need to disable interrupts around the change of %cr3, * the kernel mappings are correct, while we only grabbed the * userspace portion of VA. Interrupts handlers must not access * userspace. Having interrupts enabled fixes the issue with * firmware/SMM long operation, which would negatively affect IPIs, * esp. TLB shootdown requests. */ int efi_arch_enter(void) { pmap_t curpmap; curpmap = PCPU_GET(curpmap); PMAP_LOCK_ASSERT(curpmap, MA_OWNED); /* * IPI TLB shootdown handler invltlb_pcid_handler() reloads * %cr3 from the curpmap->pm_cr3, which would disable runtime * segments mappings. Block the handler's action by setting * curpmap to impossible value. See also comment in * pmap.c:pmap_activate_sw(). */ if (pmap_pcid_enabled && !invpcid_works) PCPU_SET(curpmap, NULL); load_cr3(VM_PAGE_TO_PHYS(efi_pml4_page) | (pmap_pcid_enabled ? curpmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid : 0)); /* * If PCID is enabled, the clear CR3_PCID_SAVE bit in the loaded %cr3 * causes TLB invalidation. */ if (!pmap_pcid_enabled) invltlb(); return (0); }
void enable_vec(struct thread *td) { int msr; struct pcb *pcb; struct trapframe *tf; pcb = td->td_pcb; tf = trapframe(td); /* * Save the thread's Altivec CPU number, and set the CPU's current * vector thread */ td->td_pcb->pcb_veccpu = PCPU_GET(cpuid); PCPU_SET(vecthread, td); /* * Enable the vector unit for when the thread returns from the * exception. If this is the first time the unit has been used by * the thread, initialise the vector registers and VSCR to 0, and * set the flag to indicate that the vector unit is in use. */ tf->srr1 |= PSL_VEC; if (!(pcb->pcb_flags & PCB_VEC)) { memset(&pcb->pcb_vec, 0, sizeof pcb->pcb_vec); pcb->pcb_flags |= PCB_VEC; } /* * Temporarily enable the vector unit so the registers * can be restored. */ msr = mfmsr(); mtmsr(msr | PSL_VEC); isync(); /* * Restore VSCR by first loading it into a vector and then into VSCR. * (this needs to done before loading the user's vector registers * since we need to use a scratch vector register) */ __asm __volatile("vxor 0,0,0; lvewx 0,0,%0; mtvscr 0" \ :: "b"(&pcb->pcb_vec.vscr)); #define LVX(n) __asm ("lvx " #n ",0,%0" \ :: "b"(&pcb->pcb_vec.vr[n])); LVX(0); LVX(1); LVX(2); LVX(3); LVX(4); LVX(5); LVX(6); LVX(7); LVX(8); LVX(9); LVX(10); LVX(11); LVX(12); LVX(13); LVX(14); LVX(15); LVX(16); LVX(17); LVX(18); LVX(19); LVX(20); LVX(21); LVX(22); LVX(23); LVX(24); LVX(25); LVX(26); LVX(27); LVX(28); LVX(29); LVX(30); LVX(31); #undef LVX isync(); mtmsr(msr); }
/* * Initialize per cpu data structures, include curthread. */ void mips_pcpu0_init() { /* Initialize pcpu info of cpu-zero */ pcpu_init(PCPU_ADDR(0), 0, sizeof(struct pcpu)); PCPU_SET(curthread, &thread0); }
/* * Initialize mips and configure to run kernel */ void mips_proc0_init(void) { #ifdef SMP if (platform_processor_id() != 0) panic("BSP must be processor number 0"); #endif proc_linkup0(&proc0, &thread0); KASSERT((kstack0 & PAGE_MASK) == 0, ("kstack0 is not aligned on a page boundary: 0x%0lx", (long)kstack0)); thread0.td_kstack = kstack0; thread0.td_kstack_pages = KSTACK_PAGES; /* * Do not use cpu_thread_alloc to initialize these fields * thread0 is the only thread that has kstack located in KSEG0 * while cpu_thread_alloc handles kstack allocated in KSEG2. */ thread0.td_pcb = (struct pcb *)(thread0.td_kstack + thread0.td_kstack_pages * PAGE_SIZE) - 1; thread0.td_frame = &thread0.td_pcb->pcb_regs; /* Steal memory for the dynamic per-cpu area. */ dpcpu_init((void *)pmap_steal_memory(DPCPU_SIZE), 0); PCPU_SET(curpcb, thread0.td_pcb); /* * There is no need to initialize md_upte array for thread0 as it's * located in .bss section and should be explicitly zeroed during * kernel initialization. */ }
static u_int ia64_ih_clock(struct thread *td, u_int xiv, struct trapframe *tf) { struct eventtimer *et; uint64_t itc, load; uint32_t mode; PCPU_INC(md.stats.pcs_nclks); intrcnt[INTRCNT_CLOCK]++; itc = ia64_get_itc(); PCPU_SET(md.clock, itc); mode = PCPU_GET(md.clock_mode); if (mode == CLOCK_ET_PERIODIC) { load = PCPU_GET(md.clock_load); ia64_set_itm(itc + load); } else ia64_set_itv((1 << 16) | xiv); ia64_set_eoi(0); ia64_srlz_d(); et = &ia64_clock_et; if (et->et_active) et->et_event_cb(et, et->et_arg); return (1); }
int set_fpregs(struct thread *td, struct fpreg *fpregs) { if (PCPU_GET(fpcurthread) == td) PCPU_SET(fpcurthread, (struct thread *)0); memcpy(&td->td_frame->f0, fpregs, sizeof(struct fpreg)); return 0; }
int acpi_wakeup_machdep(struct acpi_softc *sc, int state, int sleep_result, int intr_enabled) { if (sleep_result == -1) return (sleep_result); if (!intr_enabled) { /* Wakeup MD procedures in interrupt disabled context */ if (sleep_result == 1) { pmap_init_pat(); initializecpu(); PCPU_SET(switchtime, 0); PCPU_SET(switchticks, ticks); #ifdef DEV_APIC lapic_xapic_mode(); #endif #ifdef SMP if (!CPU_EMPTY(&suspcpus)) acpi_wakeup_cpus(sc); #endif } #ifdef SMP if (!CPU_EMPTY(&suspcpus)) restart_cpus(suspcpus); #endif mca_resume(); #ifdef __amd64__ if (vmm_resume_p != NULL) vmm_resume_p(); #endif intr_resume(/*suspend_cancelled*/false); AcpiSetFirmwareWakingVector(0, 0); } else { /* Wakeup MD procedures in interrupt enabled context */ if (sleep_result == 1 && mem_range_softc.mr_op != NULL && mem_range_softc.mr_op->reinit != NULL) mem_range_softc.mr_op->reinit(&mem_range_softc); } return (sleep_result); }
/* * Event timer start method. */ static int ia64_clock_start(struct eventtimer *et, sbintime_t first, sbintime_t period) { u_long itc, load; register_t is; if (period != 0) { PCPU_SET(md.clock_mode, CLOCK_ET_PERIODIC); load = (et->et_frequency * period) >> 32; } else {
int ia64_highfp_enable(struct thread *td, struct trapframe *tf) { struct pcb *pcb; struct pcpu *cpu; struct thread *td1; pcb = td->td_pcb; mtx_lock_spin(&ia64_highfp_mtx); cpu = pcb->pcb_fpcpu; #ifdef SMP if (cpu != NULL && cpu != pcpup) { KASSERT(cpu->pc_fpcurthread == td, ("cpu->pc_fpcurthread != td")); ia64_highfp_ipi(cpu); } #endif td1 = PCPU_GET(fpcurthread); if (td1 != NULL && td1 != td) { KASSERT(td1->td_pcb->pcb_fpcpu == pcpup, ("td1->td_pcb->pcb_fpcpu != pcpup")); save_high_fp(&td1->td_pcb->pcb_high_fp); td1->td_frame->tf_special.psr |= IA64_PSR_DFH; td1->td_pcb->pcb_fpcpu = NULL; PCPU_SET(fpcurthread, NULL); td1 = NULL; } if (td1 == NULL) { KASSERT(pcb->pcb_fpcpu == NULL, ("pcb->pcb_fpcpu != NULL")); KASSERT(PCPU_GET(fpcurthread) == NULL, ("PCPU_GET(fpcurthread) != NULL")); restore_high_fp(&pcb->pcb_high_fp); PCPU_SET(fpcurthread, td); pcb->pcb_fpcpu = pcpup; tf->tf_special.psr &= ~IA64_PSR_MFH; } tf->tf_special.psr &= ~IA64_PSR_DFH; mtx_unlock_spin(&ia64_highfp_mtx); return ((td1 != NULL) ? 1 : 0); }
noreturn void sched_run() { thread_t *td = thread_self(); PCPU_SET(idle_thread, td); td->td_slice = 0; sched_active = true; while (true) { td->td_flags |= TDF_NEEDSWITCH; } }
int acpi_wakeup_machdep(struct acpi_softc *sc, int state, int sleep_result, int intr_enabled) { if (sleep_result == -1) return (sleep_result); if (intr_enabled == 0) { /* Wakeup MD procedures in interrupt disabled context */ if (sleep_result == 1) { pmap_init_pat(); load_cr3(susppcbs[0]->pcb_cr3); initializecpu(); PCPU_SET(switchtime, 0); PCPU_SET(switchticks, ticks); #ifdef SMP if (!CPU_EMPTY(&suspcpus)) acpi_wakeup_cpus(sc, &suspcpus); #endif } #ifdef SMP if (!CPU_EMPTY(&suspcpus)) restart_cpus(suspcpus); #endif mca_resume(); intr_resume(); } else { /* Wakeup MD procedures in interrupt enabled context */ AcpiSetFirmwareWakingVector(0); if (sleep_result == 1 && mem_range_softc.mr_op != NULL && mem_range_softc.mr_op->reinit != NULL) mem_range_softc.mr_op->reinit(&mem_range_softc); } return (sleep_result); }
void efi_arch_leave(void) { pmap_t curpmap; curpmap = &curproc->p_vmspace->vm_pmap; if (pmap_pcid_enabled && !invpcid_works) PCPU_SET(curpmap, curpmap); load_cr3(curpmap->pm_cr3 | (pmap_pcid_enabled ? curpmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid : 0)); if (!pmap_pcid_enabled) invltlb(); }
static void thread_bootstrap() { thread_init(); /* Create main kernel thread */ thread_t *td = thread_create("kernel-main", (void *)kernel_init, NULL); exc_frame_t *kframe = td->td_kframe; kframe->a0 = (reg_t)_kenv.argc; kframe->a1 = (reg_t)_kenv.argv; kframe->sr |= SR_IE; /* the thread will run with interrupts enabled */ td->td_state = TDS_RUNNING; PCPU_SET(curthread, td); }
/* * Handle the return of a child process from fork1(). This function * is called from the MD fork_trampoline() entry point. */ void fork_exit(void (*callout)(void *, struct trapframe *), void *arg, struct trapframe *frame) { struct proc *p; struct thread *td; struct thread *dtd; td = curthread; p = td->td_proc; KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new")); CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)", td, td_get_sched(td), p->p_pid, td->td_name); sched_fork_exit(td); /* * Processes normally resume in mi_switch() after being * cpu_switch()'ed to, but when children start up they arrive here * instead, so we must do much the same things as mi_switch() would. */ if ((dtd = PCPU_GET(deadthread))) { PCPU_SET(deadthread, NULL); thread_stash(dtd); } thread_unlock(td); /* * cpu_fork_kthread_handler intercepts this function call to * have this call a non-return function to stay in kernel mode. * initproc has its own fork handler, but it does return. */ KASSERT(callout != NULL, ("NULL callout in fork_exit")); callout(arg, frame); /* * Check if a kernel thread misbehaved and returned from its main * function. */ if (p->p_flag & P_KPROC) { printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n", td->td_name, p->p_pid); kthread_exit(); } mtx_assert(&Giant, MA_NOTOWNED); if (p->p_sysent->sv_schedtail != NULL) (p->p_sysent->sv_schedtail)(td); td->td_pflags &= ~TDP_FORKING; }
void save_vec(struct thread *td) { struct pcb *pcb; pcb = td->td_pcb; save_vec_int(td); /* * Clear the current vec thread and pcb's CPU id * XXX should this be left clear to allow lazy save/restore ? */ pcb->pcb_veccpu = INT_MAX; PCPU_SET(vecthread, NULL); }
void save_vec(struct thread *td) { int msr; struct pcb *pcb; pcb = td->td_pcb; /* * Temporarily re-enable the vector unit during the save */ msr = mfmsr(); mtmsr(msr | PSL_VEC); isync(); /* * Save the vector registers and VSCR to the PCB */ #define STVX(n) __asm ("stvx %1,0,%0" \ :: "b"(pcb->pcb_vec.vr[n]), "n"(n)); STVX(0); STVX(1); STVX(2); STVX(3); STVX(4); STVX(5); STVX(6); STVX(7); STVX(8); STVX(9); STVX(10); STVX(11); STVX(12); STVX(13); STVX(14); STVX(15); STVX(16); STVX(17); STVX(18); STVX(19); STVX(20); STVX(21); STVX(22); STVX(23); STVX(24); STVX(25); STVX(26); STVX(27); STVX(28); STVX(29); STVX(30); STVX(31); #undef STVX __asm __volatile("mfvscr 0; stvewx 0,0,%0" :: "b"(&pcb->pcb_vec.vscr)); /* * Disable vector unit again */ isync(); mtmsr(msr); /* * Clear the current vec thread and pcb's CPU id * XXX should this be left clear to allow lazy save/restore ? */ pcb->pcb_veccpu = INT_MAX; PCPU_SET(vecthread, NULL); }
int ia64_highfp_save_ipi(void) { struct thread *td; mtx_lock_spin(&ia64_highfp_mtx); td = PCPU_GET(fpcurthread); if (td != NULL) { KASSERT(td->td_pcb->pcb_fpcpu == pcpup, ("td->td_pcb->pcb_fpcpu != pcpup")); save_high_fp(&td->td_pcb->pcb_high_fp); td->td_frame->tf_special.psr |= IA64_PSR_DFH; td->td_pcb->pcb_fpcpu = NULL; PCPU_SET(fpcurthread, NULL); } mtx_unlock_spin(&ia64_highfp_mtx); wakeup(&PCPU_GET(fpcurthread)); return ((td != NULL) ? 1 : 0); }
void cpu_mp_start(void) { int i, cpuid; mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN); cpuid = 1; for (i = 0; i < MAXCPU; i++) { if (i == boot_cpu_id) continue; if (smp_start_secondary(i)) { all_cpus |= (1 << cpuid); mp_ncpus++; cpuid++; } } idle_mask |= CR_INT_IPI; PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); }
void * initarm(void *arg, void *arg2) { struct pcpu *pc; struct pv_addr kernel_l1pt; struct pv_addr md_addr; struct pv_addr md_bla; struct pv_addr dpcpu; int loop; u_int l1pagetable; vm_offset_t freemempos; vm_offset_t lastalloced; vm_offset_t lastaddr; uint32_t memsize = 32 * 1024 * 1024; sa1110_uart_vaddr = SACOM1_VBASE; boothowto = RB_VERBOSE | RB_SINGLE; cninit(); set_cpufuncs(); lastaddr = fake_preload_metadata(); physmem = memsize / PAGE_SIZE; pc = &__pcpu; pcpu_init(pc, 0, sizeof(struct pcpu)); PCPU_SET(curthread, &thread0); /* Do basic tuning, hz etc */ init_param1(); physical_start = (vm_offset_t) KERNBASE; physical_end = lastaddr; physical_freestart = (((vm_offset_t)physical_end) + PAGE_MASK) & ~PAGE_MASK; md_addr.pv_va = md_addr.pv_pa = MDROOT_ADDR; freemempos = (vm_offset_t)round_page(physical_freestart); memset((void *)freemempos, 0, 256*1024); /* Define a macro to simplify memory allocation */ #define valloc_pages(var, np) \ alloc_pages((var).pv_pa, (np)); \ (var).pv_va = (var).pv_pa; #define alloc_pages(var, np) \ (var) = freemempos; \ freemempos += ((np) * PAGE_SIZE);\ memset((char *)(var), 0, ((np) * PAGE_SIZE)); while ((freemempos & (L1_TABLE_SIZE - 1)) != 0) freemempos += PAGE_SIZE; valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE); valloc_pages(md_bla, L2_TABLE_SIZE / PAGE_SIZE); alloc_pages(sa1_cache_clean_addr, CPU_SA110_CACHE_CLEAN_SIZE / PAGE_SIZE); for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) { if (!(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) { valloc_pages(kernel_pt_table[loop], L2_TABLE_SIZE / PAGE_SIZE); } else { kernel_pt_table[loop].pv_pa = freemempos + (loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL)) * L2_TABLE_SIZE_REAL; kernel_pt_table[loop].pv_va = kernel_pt_table[loop].pv_pa; } } /* * Allocate a page for the system page mapped to V0x00000000 * This page will just contain the system vectors and can be * shared by all processes. */ valloc_pages(systempage, 1); /* Allocate dynamic per-cpu area. */ valloc_pages(dpcpu, DPCPU_SIZE / PAGE_SIZE); dpcpu_init((void *)dpcpu.pv_va, 0); /* Allocate stacks for all modes */ valloc_pages(irqstack, IRQ_STACK_SIZE); valloc_pages(abtstack, ABT_STACK_SIZE); valloc_pages(undstack, UND_STACK_SIZE); valloc_pages(kernelstack, KSTACK_PAGES); lastalloced = kernelstack.pv_va; /* * Allocate memory for the l1 and l2 page tables. The scheme to avoid * wasting memory by allocating the l1pt on the first 16k memory was * taken from NetBSD rpc_machdep.c. NKPT should be greater than 12 for * this to work (which is supposed to be the case). */ /* * Now we start construction of the L1 page table * We start by mapping the L2 page tables into the L1. * This means that we can replace L1 mappings later on if necessary */ l1pagetable = kernel_l1pt.pv_pa; /* Map the L2 pages tables in the L1 page table */ pmap_link_l2pt(l1pagetable, 0x00000000, &kernel_pt_table[KERNEL_PT_SYS]); pmap_link_l2pt(l1pagetable, KERNBASE, &kernel_pt_table[KERNEL_PT_KERNEL]); pmap_link_l2pt(l1pagetable, 0xd0000000, &kernel_pt_table[KERNEL_PT_IO]); pmap_link_l2pt(l1pagetable, lastalloced & ~((L1_S_SIZE * 4) - 1), &kernel_pt_table[KERNEL_PT_L1]); pmap_link_l2pt(l1pagetable, 0x90000000, &kernel_pt_table[KERNEL_PT_IRQ]); pmap_link_l2pt(l1pagetable, MDROOT_ADDR, &md_bla); for (loop = 0; loop < KERNEL_PT_VMDATA_NUM; ++loop) pmap_link_l2pt(l1pagetable, KERNEL_VM_BASE + loop * 0x00100000, &kernel_pt_table[KERNEL_PT_VMDATA + loop]); pmap_map_chunk(l1pagetable, KERNBASE, KERNBASE, ((uint32_t)lastaddr - KERNBASE), VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); /* Map the DPCPU pages */ pmap_map_chunk(l1pagetable, dpcpu.pv_va, dpcpu.pv_pa, DPCPU_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); /* Map the stack pages */ pmap_map_chunk(l1pagetable, irqstack.pv_va, irqstack.pv_pa, IRQ_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, md_addr.pv_va, md_addr.pv_pa, MD_ROOT_SIZE * 1024, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, abtstack.pv_va, abtstack.pv_pa, ABT_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, undstack.pv_va, undstack.pv_pa, UND_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, kernelstack.pv_va, kernelstack.pv_pa, KSTACK_PAGES * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, kernel_l1pt.pv_va, kernel_l1pt.pv_pa, L1_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE); for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) { pmap_map_chunk(l1pagetable, kernel_pt_table[loop].pv_va, kernel_pt_table[loop].pv_pa, L2_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE); } pmap_map_chunk(l1pagetable, md_bla.pv_va, md_bla.pv_pa, L2_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE); /* Map the vector page. */ pmap_map_entry(l1pagetable, vector_page, systempage.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); /* Map the statically mapped devices. */ pmap_devmap_bootstrap(l1pagetable, assabet_devmap); pmap_map_chunk(l1pagetable, sa1_cache_clean_addr, 0xf0000000, CPU_SA110_CACHE_CLEAN_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); data_abort_handler_address = (u_int)data_abort_handler; prefetch_abort_handler_address = (u_int)prefetch_abort_handler; undefined_handler_address = (u_int)undefinedinstruction_bounce; undefined_init(); cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT); setttb(kernel_l1pt.pv_pa); cpu_tlb_flushID(); cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)); /* * Pages were allocated during the secondary bootstrap for the * stacks for different CPU modes. * We must now set the r13 registers in the different CPU modes to * point to these stacks. * Since the ARM stacks use STMFD etc. we must set r13 to the top end * of the stack memory. */ set_stackptr(PSR_IRQ32_MODE, irqstack.pv_va + IRQ_STACK_SIZE * PAGE_SIZE); set_stackptr(PSR_ABT32_MODE, abtstack.pv_va + ABT_STACK_SIZE * PAGE_SIZE); set_stackptr(PSR_UND32_MODE, undstack.pv_va + UND_STACK_SIZE * PAGE_SIZE); /* * We must now clean the cache again.... * Cleaning may be done by reading new data to displace any * dirty data in the cache. This will have happened in setttb() * but since we are boot strapping the addresses used for the read * may have just been remapped and thus the cache could be out * of sync. A re-clean after the switch will cure this. * After booting there are no gross relocations of the kernel thus * this problem will not occur after initarm(). */ cpu_idcache_wbinv_all(); bootverbose = 1; /* Set stack for exception handlers */ proc_linkup0(&proc0, &thread0); thread0.td_kstack = kernelstack.pv_va; thread0.td_pcb = (struct pcb *) (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; thread0.td_pcb->pcb_flags = 0; thread0.td_frame = &proc0_tf; /* Enable MMU, I-cache, D-cache, write buffer. */ cpufunc_control(0x337f, 0x107d); arm_vector_init(ARM_VECTORS_LOW, ARM_VEC_ALL); pmap_curmaxkvaddr = freemempos + KERNEL_PT_VMDATA_NUM * 0x400000; dump_avail[0] = phys_avail[0] = round_page(virtual_avail); dump_avail[1] = phys_avail[1] = 0xc0000000 + 0x02000000 - 1; dump_avail[2] = phys_avail[2] = 0; dump_avail[3] = phys_avail[3] = 0; mutex_init(); pmap_bootstrap(freemempos, 0xd0000000, &kernel_l1pt); init_param2(physmem); kdb_init(); return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP - sizeof(struct pcb))); }
void * initarm(struct arm_boot_params *abp) { struct pv_addr kernel_l1pt; struct pv_addr dpcpu; int loop, i; u_int l1pagetable; vm_offset_t freemempos; vm_offset_t freemem_pt; vm_offset_t afterkern; vm_offset_t freemem_after; vm_offset_t lastaddr; uint32_t memsize, memstart; lastaddr = parse_boot_param(abp); arm_physmem_kernaddr = abp->abp_physaddr; set_cpufuncs(); pcpu_init(pcpup, 0, sizeof(struct pcpu)); PCPU_SET(curthread, &thread0); /* Do basic tuning, hz etc */ init_param1(); freemempos = 0xa0200000; /* Define a macro to simplify memory allocation */ #define valloc_pages(var, np) \ alloc_pages((var).pv_pa, (np)); \ (var).pv_va = (var).pv_pa + 0x20000000; #define alloc_pages(var, np) \ freemempos -= (np * PAGE_SIZE); \ (var) = freemempos; \ memset((char *)(var), 0, ((np) * PAGE_SIZE)); while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0) freemempos -= PAGE_SIZE; valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE); for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) { if (!(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) { valloc_pages(kernel_pt_table[loop], L2_TABLE_SIZE / PAGE_SIZE); } else { kernel_pt_table[loop].pv_pa = freemempos + (loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL)) * L2_TABLE_SIZE_REAL; kernel_pt_table[loop].pv_va = kernel_pt_table[loop].pv_pa + 0x20000000; } } freemem_pt = freemempos; freemempos = 0xa0100000; /* * Allocate a page for the system page mapped to V0x00000000 * This page will just contain the system vectors and can be * shared by all processes. */ valloc_pages(systempage, 1); /* Allocate dynamic per-cpu area. */ valloc_pages(dpcpu, DPCPU_SIZE / PAGE_SIZE); dpcpu_init((void *)dpcpu.pv_va, 0); /* Allocate stacks for all modes */ valloc_pages(irqstack, IRQ_STACK_SIZE); valloc_pages(abtstack, ABT_STACK_SIZE); valloc_pages(undstack, UND_STACK_SIZE); valloc_pages(kernelstack, KSTACK_PAGES); alloc_pages(minidataclean.pv_pa, 1); valloc_pages(msgbufpv, round_page(msgbufsize) / PAGE_SIZE); /* * Allocate memory for the l1 and l2 page tables. The scheme to avoid * wasting memory by allocating the l1pt on the first 16k memory was * taken from NetBSD rpc_machdep.c. NKPT should be greater than 12 for * this to work (which is supposed to be the case). */ /* * Now we start construction of the L1 page table * We start by mapping the L2 page tables into the L1. * This means that we can replace L1 mappings later on if necessary */ l1pagetable = kernel_l1pt.pv_va; /* Map the L2 pages tables in the L1 page table */ pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH & ~(0x00100000 - 1), &kernel_pt_table[KERNEL_PT_SYS]); pmap_link_l2pt(l1pagetable, IQ80321_IOPXS_VBASE, &kernel_pt_table[KERNEL_PT_IOPXS]); pmap_link_l2pt(l1pagetable, KERNBASE, &kernel_pt_table[KERNEL_PT_BEFOREKERN]); pmap_map_chunk(l1pagetable, KERNBASE, IQ80321_SDRAM_START, 0x100000, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, KERNBASE + 0x100000, IQ80321_SDRAM_START + 0x100000, 0x100000, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE); pmap_map_chunk(l1pagetable, KERNBASE + 0x200000, IQ80321_SDRAM_START + 0x200000, (((uint32_t)(lastaddr) - KERNBASE - 0x200000) + L1_S_SIZE) & ~(L1_S_SIZE - 1), VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); freemem_after = ((int)lastaddr + PAGE_SIZE) & ~(PAGE_SIZE - 1); afterkern = round_page(((vm_offset_t)lastaddr + L1_S_SIZE) & ~(L1_S_SIZE - 1)); for (i = 0; i < KERNEL_PT_AFKERNEL_NUM; i++) { pmap_link_l2pt(l1pagetable, afterkern + i * 0x00100000, &kernel_pt_table[KERNEL_PT_AFKERNEL + i]); } pmap_map_entry(l1pagetable, afterkern, minidataclean.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); /* Map the Mini-Data cache clean area. */ xscale_setup_minidata(l1pagetable, afterkern, minidataclean.pv_pa); /* Map the vector page. */ pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); arm_devmap_bootstrap(l1pagetable, ep80219_devmap); /* * Give the XScale global cache clean code an appropriately * sized chunk of unmapped VA space starting at 0xff000000 * (our device mappings end before this address). */ xscale_cache_clean_addr = 0xff000000U; cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT); setttb(kernel_l1pt.pv_pa); cpu_tlb_flushID(); cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)); /* * Pages were allocated during the secondary bootstrap for the * stacks for different CPU modes. * We must now set the r13 registers in the different CPU modes to * point to these stacks. * Since the ARM stacks use STMFD etc. we must set r13 to the top end * of the stack memory. */ set_stackptrs(0); /* * We must now clean the cache again.... * Cleaning may be done by reading new data to displace any * dirty data in the cache. This will have happened in setttb() * but since we are boot strapping the addresses used for the read * may have just been remapped and thus the cache could be out * of sync. A re-clean after the switch will cure this. * After booting there are no gross relocations of the kernel thus * this problem will not occur after initarm(). */ cpu_idcache_wbinv_all(); cpu_setup(""); /* * Fetch the SDRAM start/size from the i80321 SDRAM configration * registers. */ i80321_calibrate_delay(); i80321_sdram_bounds(obio_bs_tag, IQ80321_80321_VBASE + VERDE_MCU_BASE, &memstart, &memsize); physmem = memsize / PAGE_SIZE; cninit(); undefined_init(); init_proc0(kernelstack.pv_va); /* Enable MMU, I-cache, D-cache, write buffer. */ arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL); vm_max_kernel_address = 0xd0000000; pmap_bootstrap(pmap_curmaxkvaddr, &kernel_l1pt); msgbufp = (void*)msgbufpv.pv_va; msgbufinit(msgbufp, msgbufsize); mutex_init(); /* * Add the physical ram we have available. * * Exclude the kernel (and all the things we allocated which immediately * follow the kernel) from the VM allocation pool but not from crash * dumps. virtual_avail is a global variable which tracks the kva we've * "allocated" while setting up pmaps. * * Prepare the list of physical memory available to the vm subsystem. */ arm_physmem_hardware_region(IQ80321_SDRAM_START, memsize); arm_physmem_exclude_region(abp->abp_physaddr, virtual_avail - KERNVIRTADDR, EXFLAG_NOALLOC); arm_physmem_init_kernel_globals(); init_param2(physmem); kdb_init(); return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP - sizeof(struct pcb))); }
void smp_init_secondary(u_int32_t cpuid) { if (cpuid >= MAXCPU) panic ("cpu id exceeds MAXCPU\n"); /* tlb init */ R4K_SetWIRED(0); R4K_TLBFlush(num_tlbentries); R4K_SetWIRED(VMWIRED_ENTRIES); MachSetPID(0); Mips_SyncCache(); mips_cp0_status_write(0); while (!aps_ready) ; mips_sync(); mips_sync(); /* Initialize curthread. */ KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread")); PCPU_SET(curthread, PCPU_GET(idlethread)); mtx_lock_spin(&ap_boot_mtx); smp_cpus++; CTR1(KTR_SMP, "SMP: AP CPU #%d Launched", PCPU_GET(cpuid)); /* Build our map of 'other' CPUs. */ PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); printf("SMP: AP CPU #%d Launched!\n", PCPU_GET(cpuid)); if (smp_cpus == mp_ncpus) { smp_started = 1; smp_active = 1; } mtx_unlock_spin(&ap_boot_mtx); while (smp_started == 0) ; /* nothing */ /* Enable Interrupt */ mips_cp0_status_write(SR_INT_ENAB); /* ok, now grab sched_lock and enter the scheduler */ mtx_lock_spin(&sched_lock); /* * Correct spinlock nesting. The idle thread context that we are * borrowing was created so that it would start out with a single * spin lock (sched_lock) held in fork_trampoline(). Since we've * explicitly acquired locks in this function, the nesting count * is now 2 rather than 1. Since we are nested, calling * spinlock_exit() will simply adjust the counts without allowing * spin lock using code to interrupt us. */ spinlock_exit(); KASSERT(curthread->td_md.md_spinlock_count == 1, ("invalid count")); binuptime(PCPU_PTR(switchtime)); PCPU_SET(switchticks, ticks); /* kick off the clock on this cpu */ mips_start_timer(); cpu_throw(NULL, choosethread()); /* doesn't return */ panic("scheduler returned us to %s", __func__); }
/* * AP CPU's call this to initialize themselves. */ void init_secondary(void) { vm_offset_t addr; u_int cpuid; int gsel_tss; /* bootAP is set in start_ap() to our ID. */ PCPU_SET(currentldt, _default_ldt); gsel_tss = GSEL(GPROC0_SEL, SEL_KPL); #if 0 gdt[bootAP * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS; #endif PCPU_SET(common_tss.tss_esp0, 0); /* not used until after switch */ PCPU_SET(common_tss.tss_ss0, GSEL(GDATA_SEL, SEL_KPL)); PCPU_SET(common_tss.tss_ioopt, (sizeof (struct i386tss)) << 16); #if 0 PCPU_SET(tss_gdt, &gdt[bootAP * NGDT + GPROC0_SEL].sd); PCPU_SET(common_tssd, *PCPU_GET(tss_gdt)); #endif PCPU_SET(fsgs_gdt, &gdt[GUFS_SEL].sd); /* * Set to a known state: * Set by mpboot.s: CR0_PG, CR0_PE * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM */ /* * signal our startup to the BSP. */ mp_naps++; /* Spin until the BSP releases the AP's. */ while (!aps_ready) ia32_pause(); /* BSP may have changed PTD while we were waiting */ invltlb(); for (addr = 0; addr < NKPT * NBPDR - 1; addr += PAGE_SIZE) invlpg(addr); /* set up FPU state on the AP */ npxinit(); #if 0 /* set up SSE registers */ enable_sse(); #endif #if 0 && defined(PAE) /* Enable the PTE no-execute bit. */ if ((amd_feature & AMDID_NX) != 0) { uint64_t msr; msr = rdmsr(MSR_EFER) | EFER_NXE; wrmsr(MSR_EFER, msr); } #endif #if 0 /* A quick check from sanity claus */ if (PCPU_GET(apic_id) != lapic_id()) { printf("SMP: cpuid = %d\n", PCPU_GET(cpuid)); printf("SMP: actual apic_id = %d\n", lapic_id()); printf("SMP: correct apic_id = %d\n", PCPU_GET(apic_id)); panic("cpuid mismatch! boom!!"); } #endif /* Initialize curthread. */ KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread")); PCPU_SET(curthread, PCPU_GET(idlethread)); mtx_lock_spin(&ap_boot_mtx); #if 0 /* Init local apic for irq's */ lapic_setup(1); #endif smp_cpus++; cpuid = PCPU_GET(cpuid); CTR1(KTR_SMP, "SMP: AP CPU #%d Launched", cpuid); printf("SMP: AP CPU #%d Launched!\n", cpuid); /* Determine if we are a logical CPU. */ if (logical_cpus > 1 && PCPU_GET(apic_id) % logical_cpus != 0) CPU_SET(cpuid, &logical_cpus_mask); /* Determine if we are a hyperthread. */ if (hyperthreading_cpus > 1 && PCPU_GET(apic_id) % hyperthreading_cpus != 0) CPU_SET(cpuid, &hyperthreading_cpus_mask); #if 0 if (bootverbose) lapic_dump("AP"); #endif if (smp_cpus == mp_ncpus) { /* enable IPI's, tlb shootdown, freezes etc */ atomic_store_rel_int(&smp_started, 1); smp_active = 1; /* historic */ } mtx_unlock_spin(&ap_boot_mtx); /* wait until all the AP's are up */ while (smp_started == 0) ia32_pause(); PCPU_SET(curthread, PCPU_GET(idlethread)); /* Start per-CPU event timers. */ cpu_initclocks_ap(); /* enter the scheduler */ sched_throw(NULL); panic("scheduler returned us to %s", __func__); /* NOTREACHED */ }
void * initarm(void *arg, void *arg2) { #define next_chunk2(a,b) (((a) + (b)) &~ ((b)-1)) #define next_page(a) next_chunk2(a,PAGE_SIZE) struct pv_addr kernel_l1pt; struct pv_addr dpcpu; int loop, i; u_int l1pagetable; vm_offset_t freemempos; vm_offset_t freemem_pt; vm_offset_t afterkern; vm_offset_t freemem_after; vm_offset_t lastaddr; uint32_t memsize; set_cpufuncs(); /* NB: sets cputype */ lastaddr = fake_preload_metadata(); pcpu_init(pcpup, 0, sizeof(struct pcpu)); PCPU_SET(curthread, &thread0); /* Do basic tuning, hz etc */ init_param1(); /* * We allocate memory downwards from where we were loaded * by RedBoot; first the L1 page table, then NUM_KERNEL_PTS * entries in the L2 page table. Past that we re-align the * allocation boundary so later data structures (stacks, etc) * can be mapped with different attributes (write-back vs * write-through). Note this leaves a gap for expansion * (or might be repurposed). */ freemempos = KERNPHYSADDR; /* macros to simplify initial memory allocation */ #define alloc_pages(var, np) do { \ freemempos -= (np * PAGE_SIZE); \ (var) = freemempos; \ /* NB: this works because locore maps PA=VA */ \ memset((char *)(var), 0, ((np) * PAGE_SIZE)); \ } while (0) #define valloc_pages(var, np) do { \ alloc_pages((var).pv_pa, (np)); \ (var).pv_va = (var).pv_pa + (KERNVIRTADDR - KERNPHYSADDR); \ } while (0) /* force L1 page table alignment */ while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0) freemempos -= PAGE_SIZE; /* allocate contiguous L1 page table */ valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE); /* now allocate L2 page tables; they are linked to L1 below */ for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) { if (!(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) { valloc_pages(kernel_pt_table[loop], L2_TABLE_SIZE / PAGE_SIZE); } else { kernel_pt_table[loop].pv_pa = freemempos + (loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL)) * L2_TABLE_SIZE_REAL; kernel_pt_table[loop].pv_va = kernel_pt_table[loop].pv_pa + (KERNVIRTADDR - KERNPHYSADDR); } } freemem_pt = freemempos; /* base of allocated pt's */ /* * Re-align allocation boundary so we can map the area * write-back instead of write-through for the stacks and * related structures allocated below. */ freemempos = PHYSADDR + 0x100000; /* * Allocate a page for the system page mapped to V0x00000000 * This page will just contain the system vectors and can be * shared by all processes. */ valloc_pages(systempage, 1); /* Allocate dynamic per-cpu area. */ valloc_pages(dpcpu, DPCPU_SIZE / PAGE_SIZE); dpcpu_init((void *)dpcpu.pv_va, 0); /* Allocate stacks for all modes */ valloc_pages(irqstack, IRQ_STACK_SIZE); valloc_pages(abtstack, ABT_STACK_SIZE); valloc_pages(undstack, UND_STACK_SIZE); valloc_pages(kernelstack, KSTACK_PAGES); alloc_pages(minidataclean.pv_pa, 1); valloc_pages(msgbufpv, round_page(msgbufsize) / PAGE_SIZE); #ifdef ARM_USE_SMALL_ALLOC freemempos -= PAGE_SIZE; freemem_pt = trunc_page(freemem_pt); freemem_after = freemempos - ((freemem_pt - (PHYSADDR + 0x100000)) / PAGE_SIZE) * sizeof(struct arm_small_page); arm_add_smallalloc_pages( (void *)(freemem_after + (KERNVIRTADDR - KERNPHYSADDR)), (void *)0xc0100000, freemem_pt - (PHYSADDR + 0x100000), 1); freemem_after -= ((freemem_after - (PHYSADDR + 0x1000)) / PAGE_SIZE) * sizeof(struct arm_small_page); arm_add_smallalloc_pages( (void *)(freemem_after + (KERNVIRTADDR - KERNPHYSADDR)), (void *)0xc0001000, trunc_page(freemem_after) - (PHYSADDR + 0x1000), 0); freemempos = trunc_page(freemem_after); freemempos -= PAGE_SIZE; #endif /* * Now construct the L1 page table. First map the L2 * page tables into the L1 so we can replace L1 mappings * later on if necessary */ l1pagetable = kernel_l1pt.pv_va; /* Map the L2 pages tables in the L1 page table */ pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH & ~(0x00100000 - 1), &kernel_pt_table[KERNEL_PT_SYS]); pmap_link_l2pt(l1pagetable, IXP425_IO_VBASE, &kernel_pt_table[KERNEL_PT_IO]); pmap_link_l2pt(l1pagetable, IXP425_MCU_VBASE, &kernel_pt_table[KERNEL_PT_IO + 1]); pmap_link_l2pt(l1pagetable, IXP425_PCI_MEM_VBASE, &kernel_pt_table[KERNEL_PT_IO + 2]); pmap_link_l2pt(l1pagetable, KERNBASE, &kernel_pt_table[KERNEL_PT_BEFOREKERN]); pmap_map_chunk(l1pagetable, KERNBASE, PHYSADDR, 0x100000, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, KERNBASE + 0x100000, PHYSADDR + 0x100000, 0x100000, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE); pmap_map_chunk(l1pagetable, KERNEL_TEXT_BASE, KERNEL_TEXT_PHYS, next_chunk2(((uint32_t)lastaddr) - KERNEL_TEXT_BASE, L1_S_SIZE), VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); freemem_after = next_page((int)lastaddr); afterkern = round_page(next_chunk2((vm_offset_t)lastaddr, L1_S_SIZE)); for (i = 0; i < KERNEL_PT_AFKERNEL_NUM; i++) { pmap_link_l2pt(l1pagetable, afterkern + i * 0x00100000, &kernel_pt_table[KERNEL_PT_AFKERNEL + i]); } pmap_map_entry(l1pagetable, afterkern, minidataclean.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); #ifdef ARM_USE_SMALL_ALLOC if ((freemem_after + 2 * PAGE_SIZE) <= afterkern) { arm_add_smallalloc_pages((void *)(freemem_after), (void*)(freemem_after + PAGE_SIZE), afterkern - (freemem_after + PAGE_SIZE), 0); } #endif /* Map the Mini-Data cache clean area. */ xscale_setup_minidata(l1pagetable, afterkern, minidataclean.pv_pa); /* Map the vector page. */ pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); if (cpu_is_ixp43x()) pmap_devmap_bootstrap(l1pagetable, ixp435_devmap); else pmap_devmap_bootstrap(l1pagetable, ixp425_devmap); /* * Give the XScale global cache clean code an appropriately * sized chunk of unmapped VA space starting at 0xff000000 * (our device mappings end before this address). */ xscale_cache_clean_addr = 0xff000000U; cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT); setttb(kernel_l1pt.pv_pa); cpu_tlb_flushID(); cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)); /* * Pages were allocated during the secondary bootstrap for the * stacks for different CPU modes. * We must now set the r13 registers in the different CPU modes to * point to these stacks. * Since the ARM stacks use STMFD etc. we must set r13 to the top end * of the stack memory. */ set_stackptr(PSR_IRQ32_MODE, irqstack.pv_va + IRQ_STACK_SIZE*PAGE_SIZE); set_stackptr(PSR_ABT32_MODE, abtstack.pv_va + ABT_STACK_SIZE*PAGE_SIZE); set_stackptr(PSR_UND32_MODE, undstack.pv_va + UND_STACK_SIZE*PAGE_SIZE); /* * We must now clean the cache again.... * Cleaning may be done by reading new data to displace any * dirty data in the cache. This will have happened in setttb() * but since we are boot strapping the addresses used for the read * may have just been remapped and thus the cache could be out * of sync. A re-clean after the switch will cure this. * After booting there are no gross relocations of the kernel thus * this problem will not occur after initarm(). */ cpu_idcache_wbinv_all(); /* ready to setup the console (XXX move earlier if possible) */ cninit(); /* * Fetch the RAM size from the MCU registers. The * expansion bus was mapped above so we can now read 'em. */ if (cpu_is_ixp43x()) memsize = ixp435_ddram_size(); else memsize = ixp425_sdram_size(); physmem = memsize / PAGE_SIZE; /* Set stack for exception handlers */ data_abort_handler_address = (u_int)data_abort_handler; prefetch_abort_handler_address = (u_int)prefetch_abort_handler; undefined_handler_address = (u_int)undefinedinstruction_bounce; undefined_init(); proc_linkup0(&proc0, &thread0); thread0.td_kstack = kernelstack.pv_va; thread0.td_pcb = (struct pcb *) (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; thread0.td_pcb->pcb_flags = 0; thread0.td_frame = &proc0_tf; pcpup->pc_curpcb = thread0.td_pcb; arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL); pmap_curmaxkvaddr = afterkern + PAGE_SIZE; dump_avail[0] = PHYSADDR; dump_avail[1] = PHYSADDR + memsize; dump_avail[2] = 0; dump_avail[3] = 0; pmap_bootstrap(pmap_curmaxkvaddr, 0xd0000000, &kernel_l1pt); msgbufp = (void*)msgbufpv.pv_va; msgbufinit(msgbufp, msgbufsize); mutex_init(); i = 0; #ifdef ARM_USE_SMALL_ALLOC phys_avail[i++] = PHYSADDR; phys_avail[i++] = PHYSADDR + PAGE_SIZE; /* *XXX: Gross hack to get our * pages in the vm_page_array. */ #endif phys_avail[i++] = round_page(virtual_avail - KERNBASE + PHYSADDR); phys_avail[i++] = trunc_page(PHYSADDR + memsize - 1); phys_avail[i++] = 0; phys_avail[i] = 0; init_param2(physmem); kdb_init(); /* use static kernel environment if so configured */ if (envmode == 1) kern_envp = static_env; return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP - sizeof(struct pcb))); #undef next_page #undef next_chunk2 }
void initarm(struct arm64_bootparams *abp) { struct efi_map_header *efihdr; struct pcpu *pcpup; vm_offset_t lastaddr; caddr_t kmdp; vm_paddr_t mem_len; int i; /* Set the module data location */ preload_metadata = (caddr_t)(uintptr_t)(abp->modulep); /* Find the kernel address */ kmdp = preload_search_by_type("elf kernel"); if (kmdp == NULL) kmdp = preload_search_by_type("elf64 kernel"); boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int); kern_envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *); #ifdef FDT try_load_dtb(kmdp); #endif /* Find the address to start allocating from */ lastaddr = MD_FETCH(kmdp, MODINFOMD_KERNEND, vm_offset_t); /* Load the physical memory ranges */ physmap_idx = 0; efihdr = (struct efi_map_header *)preload_search_info(kmdp, MODINFO_METADATA | MODINFOMD_EFI_MAP); add_efi_map_entries(efihdr, physmap, &physmap_idx); /* Print the memory map */ mem_len = 0; for (i = 0; i < physmap_idx; i += 2) { dump_avail[i] = physmap[i]; dump_avail[i + 1] = physmap[i + 1]; mem_len += physmap[i + 1] - physmap[i]; } dump_avail[i] = 0; dump_avail[i + 1] = 0; /* Set the pcpu data, this is needed by pmap_bootstrap */ pcpup = &__pcpu[0]; pcpu_init(pcpup, 0, sizeof(struct pcpu)); /* * Set the pcpu pointer with a backup in tpidr_el1 to be * loaded when entering the kernel from userland. */ __asm __volatile( "mov x18, %0 \n" "msr tpidr_el1, %0" :: "r"(pcpup)); PCPU_SET(curthread, &thread0); /* Do basic tuning, hz etc */ init_param1(); cache_setup(); /* Bootstrap enough of pmap to enter the kernel proper */ pmap_bootstrap(abp->kern_l1pt, KERNBASE - abp->kern_delta, lastaddr - KERNBASE); arm_devmap_bootstrap(0, NULL); cninit(); init_proc0(abp->kern_stack); msgbufinit(msgbufp, msgbufsize); mutex_init(); init_param2(physmem); dbg_monitor_init(); kdb_init(); early_boot = 0; }
void * initarm(struct arm_boot_params *abp) { #define next_chunk2(a,b) (((a) + (b)) &~ ((b)-1)) #define next_page(a) next_chunk2(a,PAGE_SIZE) struct pv_addr kernel_l1pt; struct pv_addr dpcpu; int loop, i; u_int l1pagetable; vm_offset_t freemempos; vm_offset_t freemem_pt; vm_offset_t afterkern; vm_offset_t freemem_after; vm_offset_t lastaddr; uint32_t memsize; /* kernel text starts where we were loaded at boot */ #define KERNEL_TEXT_OFF (abp->abp_physaddr - PHYSADDR) #define KERNEL_TEXT_BASE (KERNBASE + KERNEL_TEXT_OFF) #define KERNEL_TEXT_PHYS (PHYSADDR + KERNEL_TEXT_OFF) lastaddr = parse_boot_param(abp); arm_physmem_kernaddr = abp->abp_physaddr; set_cpufuncs(); /* NB: sets cputype */ pcpu_init(pcpup, 0, sizeof(struct pcpu)); PCPU_SET(curthread, &thread0); if (envmode == 1) kern_envp = static_env; /* Do basic tuning, hz etc */ init_param1(); /* * We allocate memory downwards from where we were loaded * by RedBoot; first the L1 page table, then NUM_KERNEL_PTS * entries in the L2 page table. Past that we re-align the * allocation boundary so later data structures (stacks, etc) * can be mapped with different attributes (write-back vs * write-through). Note this leaves a gap for expansion * (or might be repurposed). */ freemempos = abp->abp_physaddr; /* macros to simplify initial memory allocation */ #define alloc_pages(var, np) do { \ freemempos -= (np * PAGE_SIZE); \ (var) = freemempos; \ /* NB: this works because locore maps PA=VA */ \ memset((char *)(var), 0, ((np) * PAGE_SIZE)); \ } while (0) #define valloc_pages(var, np) do { \ alloc_pages((var).pv_pa, (np)); \ (var).pv_va = (var).pv_pa + (KERNVIRTADDR - abp->abp_physaddr); \ } while (0) /* force L1 page table alignment */ while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0) freemempos -= PAGE_SIZE; /* allocate contiguous L1 page table */ valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE); /* now allocate L2 page tables; they are linked to L1 below */ for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) { if (!(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) { valloc_pages(kernel_pt_table[loop], L2_TABLE_SIZE / PAGE_SIZE); } else { kernel_pt_table[loop].pv_pa = freemempos + (loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL)) * L2_TABLE_SIZE_REAL; kernel_pt_table[loop].pv_va = kernel_pt_table[loop].pv_pa + (KERNVIRTADDR - abp->abp_physaddr); } } freemem_pt = freemempos; /* base of allocated pt's */ /* * Re-align allocation boundary so we can map the area * write-back instead of write-through for the stacks and * related structures allocated below. */ freemempos = PHYSADDR + 0x100000; /* * Allocate a page for the system page mapped to V0x00000000 * This page will just contain the system vectors and can be * shared by all processes. */ valloc_pages(systempage, 1); /* Allocate dynamic per-cpu area. */ valloc_pages(dpcpu, DPCPU_SIZE / PAGE_SIZE); dpcpu_init((void *)dpcpu.pv_va, 0); /* Allocate stacks for all modes */ valloc_pages(irqstack, IRQ_STACK_SIZE); valloc_pages(abtstack, ABT_STACK_SIZE); valloc_pages(undstack, UND_STACK_SIZE); valloc_pages(kernelstack, KSTACK_PAGES); alloc_pages(minidataclean.pv_pa, 1); valloc_pages(msgbufpv, round_page(msgbufsize) / PAGE_SIZE); /* * Now construct the L1 page table. First map the L2 * page tables into the L1 so we can replace L1 mappings * later on if necessary */ l1pagetable = kernel_l1pt.pv_va; /* Map the L2 pages tables in the L1 page table */ pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH & ~(0x00100000 - 1), &kernel_pt_table[KERNEL_PT_SYS]); pmap_link_l2pt(l1pagetable, IXP425_IO_VBASE, &kernel_pt_table[KERNEL_PT_IO]); pmap_link_l2pt(l1pagetable, IXP425_MCU_VBASE, &kernel_pt_table[KERNEL_PT_IO + 1]); pmap_link_l2pt(l1pagetable, IXP425_PCI_MEM_VBASE, &kernel_pt_table[KERNEL_PT_IO + 2]); pmap_link_l2pt(l1pagetable, KERNBASE, &kernel_pt_table[KERNEL_PT_BEFOREKERN]); pmap_map_chunk(l1pagetable, KERNBASE, PHYSADDR, 0x100000, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); pmap_map_chunk(l1pagetable, KERNBASE + 0x100000, PHYSADDR + 0x100000, 0x100000, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE); pmap_map_chunk(l1pagetable, KERNEL_TEXT_BASE, KERNEL_TEXT_PHYS, next_chunk2(((uint32_t)lastaddr) - KERNEL_TEXT_BASE, L1_S_SIZE), VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); freemem_after = next_page((int)lastaddr); afterkern = round_page(next_chunk2((vm_offset_t)lastaddr, L1_S_SIZE)); for (i = 0; i < KERNEL_PT_AFKERNEL_NUM; i++) { pmap_link_l2pt(l1pagetable, afterkern + i * 0x00100000, &kernel_pt_table[KERNEL_PT_AFKERNEL + i]); } pmap_map_entry(l1pagetable, afterkern, minidataclean.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); /* Map the Mini-Data cache clean area. */ xscale_setup_minidata(l1pagetable, afterkern, minidataclean.pv_pa); /* Map the vector page. */ pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE); if (cpu_is_ixp43x()) arm_devmap_bootstrap(l1pagetable, ixp435_devmap); else arm_devmap_bootstrap(l1pagetable, ixp425_devmap); /* * Give the XScale global cache clean code an appropriately * sized chunk of unmapped VA space starting at 0xff000000 * (our device mappings end before this address). */ xscale_cache_clean_addr = 0xff000000U; cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT); setttb(kernel_l1pt.pv_pa); cpu_tlb_flushID(); cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)); /* * Pages were allocated during the secondary bootstrap for the * stacks for different CPU modes. * We must now set the r13 registers in the different CPU modes to * point to these stacks. * Since the ARM stacks use STMFD etc. we must set r13 to the top end * of the stack memory. */ set_stackptrs(0); /* * We must now clean the cache again.... * Cleaning may be done by reading new data to displace any * dirty data in the cache. This will have happened in setttb() * but since we are boot strapping the addresses used for the read * may have just been remapped and thus the cache could be out * of sync. A re-clean after the switch will cure this. * After booting there are no gross relocations of the kernel thus * this problem will not occur after initarm(). */ cpu_idcache_wbinv_all(); cpu_setup(); /* ready to setup the console (XXX move earlier if possible) */ cninit(); /* * Fetch the RAM size from the MCU registers. The * expansion bus was mapped above so we can now read 'em. */ if (cpu_is_ixp43x()) memsize = ixp435_ddram_size(); else memsize = ixp425_sdram_size(); undefined_init(); init_proc0(kernelstack.pv_va); arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL); pmap_curmaxkvaddr = afterkern + PAGE_SIZE; vm_max_kernel_address = 0xe0000000; pmap_bootstrap(pmap_curmaxkvaddr, &kernel_l1pt); msgbufp = (void*)msgbufpv.pv_va; msgbufinit(msgbufp, msgbufsize); mutex_init(); /* * Add the physical ram we have available. * * Exclude the kernel, and all the things we allocated which immediately * follow the kernel, from the VM allocation pool but not from crash * dumps. virtual_avail is a global variable which tracks the kva we've * "allocated" while setting up pmaps. * * Prepare the list of physical memory available to the vm subsystem. */ arm_physmem_hardware_region(PHYSADDR, memsize); arm_physmem_exclude_region(freemem_pt, KERNPHYSADDR - freemem_pt, EXFLAG_NOALLOC); arm_physmem_exclude_region(freemempos, KERNPHYSADDR - 0x100000 - freemempos, EXFLAG_NOALLOC); arm_physmem_exclude_region(abp->abp_physaddr, virtual_avail - KERNVIRTADDR, EXFLAG_NOALLOC); arm_physmem_init_kernel_globals(); init_param2(physmem); kdb_init(); /* use static kernel environment if so configured */ if (envmode == 1) kern_envp = static_env; return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP - sizeof(struct pcb))); #undef next_page #undef next_chunk2 }
/* * Map the local APIC and setup necessary interrupt vectors. */ static void native_lapic_init(vm_paddr_t addr) { uint32_t ver; u_int regs[4]; int i, arat; /* * Enable x2APIC mode if possible. Map the local APIC * registers page. * * Keep the LAPIC registers page mapped uncached for x2APIC * mode too, to have direct map page attribute set to * uncached. This is needed to work around CPU errata present * on all Intel processors. */ KASSERT(trunc_page(addr) == addr, ("local APIC not aligned on a page boundary")); lapic_paddr = addr; lapic_map = pmap_mapdev(addr, PAGE_SIZE); if (x2apic_mode) { native_lapic_enable_x2apic(); lapic_map = NULL; } /* Setup the spurious interrupt handler. */ setidt(APIC_SPURIOUS_INT, IDTVEC(spuriousint), SDT_APIC, SEL_KPL, GSEL_APIC); /* Perform basic initialization of the BSP's local APIC. */ lapic_enable(); /* Set BSP's per-CPU local APIC ID. */ PCPU_SET(apic_id, lapic_id()); /* Local APIC timer interrupt. */ setidt(APIC_TIMER_INT, IDTVEC(timerint), SDT_APIC, SEL_KPL, GSEL_APIC); /* Local APIC error interrupt. */ setidt(APIC_ERROR_INT, IDTVEC(errorint), SDT_APIC, SEL_KPL, GSEL_APIC); /* XXX: Thermal interrupt */ /* Local APIC CMCI. */ setidt(APIC_CMC_INT, IDTVEC(cmcint), SDT_APICT, SEL_KPL, GSEL_APIC); if ((resource_int_value("apic", 0, "clock", &i) != 0 || i != 0)) { arat = 0; /* Intel CPUID 0x06 EAX[2] set if APIC timer runs in C3. */ if (cpu_vendor_id == CPU_VENDOR_INTEL && cpu_high >= 6) { do_cpuid(0x06, regs); if ((regs[0] & CPUTPM1_ARAT) != 0) arat = 1; } bzero(&lapic_et, sizeof(lapic_et)); lapic_et.et_name = "LAPIC"; lapic_et.et_flags = ET_FLAGS_PERIODIC | ET_FLAGS_ONESHOT | ET_FLAGS_PERCPU; lapic_et.et_quality = 600; if (!arat) { lapic_et.et_flags |= ET_FLAGS_C3STOP; lapic_et.et_quality -= 200; } lapic_et.et_frequency = 0; /* We don't know frequency yet, so trying to guess. */ lapic_et.et_min_period = 0x00001000LL; lapic_et.et_max_period = SBT_1S; lapic_et.et_start = lapic_et_start; lapic_et.et_stop = lapic_et_stop; lapic_et.et_priv = NULL; et_register(&lapic_et); } /* * Set lapic_eoi_suppression after lapic_enable(), to not * enable suppression in the hardware prematurely. Note that * we by default enable suppression even when system only has * one IO-APIC, since EOI is broadcasted to all APIC agents, * including CPUs, otherwise. */ ver = lapic_read32(LAPIC_VERSION); if ((ver & APIC_VER_EOI_SUPPRESSION) != 0) { lapic_eoi_suppression = 1; TUNABLE_INT_FETCH("hw.lapic_eoi_suppression", &lapic_eoi_suppression); } }
static void cheriabi_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack) { struct cheri_signal *csigp; u_long stackbase, stacklen; bzero((caddr_t)td->td_frame, sizeof(struct trapframe)); KASSERT(stack % sizeof(struct chericap) == 0, ("CheriABI stack pointer not properly aligned")); cheri_capability_set(&td->td_proc->p_md.md_cheri_mmap_cap, CHERI_CAP_USER_MMAP_PERMS, CHERI_CAP_USER_MMAP_BASE, CHERI_CAP_USER_MMAP_LENGTH, CHERI_CAP_USER_MMAP_OFFSET); td->td_frame->pc = imgp->entry_addr; td->td_frame->sr = MIPS_SR_KSU_USER | MIPS_SR_EXL | MIPS_SR_INT_IE | (mips_rd_status() & MIPS_SR_INT_MASK) | MIPS_SR_PX | MIPS_SR_UX | MIPS_SR_KX | MIPS_SR_COP_2_BIT; cheri_exec_setregs(td, imgp->entry_addr); cheri_stack_init(td->td_pcb); /* * Pass a pointer to the struct cheriabi_execdata at the top of the * stack. * * XXXBD: should likely be read only */ cheri_capability_set(&td->td_frame->c3, CHERI_CAP_USER_DATA_PERMS, (void *)stack, sizeof(struct cheriabi_execdata), 0); /* * Restrict the stack capability to the maximum region allowed for * this process and adjust sp accordingly. * * XXXBD: 8MB should be the process stack limit. */ CTASSERT(CHERI_CAP_USER_DATA_BASE == 0); stackbase = USRSTACK - (1024 * 1024 * 8); KASSERT(stack > stackbase, ("top of stack 0x%lx is below stack base 0x%lx", stack, stackbase)); stacklen = stack - stackbase; cheri_capability_set(&td->td_frame->stc, CHERI_CAP_USER_DATA_PERMS, (void *)stackbase, stacklen, 0); td->td_frame->sp = stacklen; /* * Also update the signal stack. The default set in * cheri_exec_setregs() covers the whole address space. */ csigp = &td->td_pcb->pcb_cherisignal; cheri_capability_set(&csigp->csig_stc, CHERI_CAP_USER_DATA_PERMS, (void *)stackbase, stacklen, 0); /* XXX: set sp for signal stack! */ td->td_md.md_flags &= ~MDTD_FPUSED; if (PCPU_GET(fpcurthread) == td) PCPU_SET(fpcurthread, (struct thread *)0); td->td_md.md_ss_addr = 0; td->td_md.md_tls_tcb_offset = TLS_TP_OFFSET + TLS_TCB_SIZE_C; }