Index: head/sys/arm64/arm64/identcpu.c =================================================================== --- head/sys/arm64/arm64/identcpu.c (revision 307209) +++ head/sys/arm64/arm64/identcpu.c (revision 307210) @@ -1,787 +1,800 @@ /*- * Copyright (c) 2014 Andrew Turner * Copyright (c) 2014 The FreeBSD Foundation * All rights reserved. * * Portions of this software were developed by Semihalf * under sponsorship of the FreeBSD Foundation. * * 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 +#include #include +#include #include #include #include #include #include static int ident_lock; char machine[] = "arm64"; SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD, machine, 0, "Machine class"); /* * Per-CPU affinity as provided in MPIDR_EL1 * Indexed by CPU number in logical order selected by the system. * Relevant fields can be extracted using CPU_AFFn macros, * Aff3.Aff2.Aff1.Aff0 construct a unique CPU address in the system. * * Fields used by us: * Aff1 - Cluster number * Aff0 - CPU number in Aff1 cluster */ uint64_t __cpu_affinity[MAXCPU]; static u_int cpu_aff_levels; struct cpu_desc { u_int cpu_impl; u_int cpu_part_num; u_int cpu_variant; u_int cpu_revision; const char *cpu_impl_name; const char *cpu_part_name; uint64_t mpidr; uint64_t id_aa64afr0; uint64_t id_aa64afr1; uint64_t id_aa64dfr0; uint64_t id_aa64dfr1; uint64_t id_aa64isar0; uint64_t id_aa64isar1; uint64_t id_aa64mmfr0; uint64_t id_aa64mmfr1; uint64_t id_aa64pfr0; uint64_t id_aa64pfr1; }; struct cpu_desc cpu_desc[MAXCPU]; static u_int cpu_print_regs; #define PRINT_ID_AA64_AFR0 0x00000001 #define PRINT_ID_AA64_AFR1 0x00000002 #define PRINT_ID_AA64_DFR0 0x00000004 #define PRINT_ID_AA64_DFR1 0x00000008 #define PRINT_ID_AA64_ISAR0 0x00000010 #define PRINT_ID_AA64_ISAR1 0x00000020 #define PRINT_ID_AA64_MMFR0 0x00000040 #define PRINT_ID_AA64_MMFR1 0x00000080 #define PRINT_ID_AA64_PFR0 0x00000100 #define PRINT_ID_AA64_PFR1 0x00000200 struct cpu_parts { u_int part_id; const char *part_name; }; #define CPU_PART_NONE { 0, "Unknown Processor" } struct cpu_implementers { u_int impl_id; const char *impl_name; /* * Part number is implementation defined * so each vendor will have its own set of values and names. */ const struct cpu_parts *cpu_parts; }; #define CPU_IMPLEMENTER_NONE { 0, "Unknown Implementer", cpu_parts_none } /* * Per-implementer table of (PartNum, CPU Name) pairs. */ /* ARM Ltd. */ static const struct cpu_parts cpu_parts_arm[] = { { CPU_PART_FOUNDATION, "Foundation-Model" }, { CPU_PART_CORTEX_A53, "Cortex-A53" }, { CPU_PART_CORTEX_A57, "Cortex-A57" }, CPU_PART_NONE, }; /* Cavium */ static const struct cpu_parts cpu_parts_cavium[] = { { CPU_PART_THUNDER, "Thunder" }, CPU_PART_NONE, }; /* Unknown */ static const struct cpu_parts cpu_parts_none[] = { CPU_PART_NONE, }; /* * Implementers table. */ const struct cpu_implementers cpu_implementers[] = { { CPU_IMPL_ARM, "ARM", cpu_parts_arm }, { CPU_IMPL_BROADCOM, "Broadcom", cpu_parts_none }, { CPU_IMPL_CAVIUM, "Cavium", cpu_parts_cavium }, { CPU_IMPL_DEC, "DEC", cpu_parts_none }, { CPU_IMPL_INFINEON, "IFX", cpu_parts_none }, { CPU_IMPL_FREESCALE, "Freescale", cpu_parts_none }, { CPU_IMPL_NVIDIA, "NVIDIA", cpu_parts_none }, { CPU_IMPL_APM, "APM", cpu_parts_none }, { CPU_IMPL_QUALCOMM, "Qualcomm", cpu_parts_none }, { CPU_IMPL_MARVELL, "Marvell", cpu_parts_none }, { CPU_IMPL_INTEL, "Intel", cpu_parts_none }, CPU_IMPLEMENTER_NONE, }; + +static void +identify_cpu_sysinit(void *dummy __unused) +{ + int cpu; + + CPU_FOREACH(cpu) { + print_cpu_features(cpu); + } +} +SYSINIT(idenrity_cpu, SI_SUB_SMP, SI_ORDER_ANY, identify_cpu_sysinit, NULL); void print_cpu_features(u_int cpu) { int printed; printf("CPU%3d: %s %s r%dp%d", cpu, cpu_desc[cpu].cpu_impl_name, cpu_desc[cpu].cpu_part_name, cpu_desc[cpu].cpu_variant, cpu_desc[cpu].cpu_revision); printf(" affinity:"); switch(cpu_aff_levels) { default: case 4: printf(" %2d", CPU_AFF3(cpu_desc[cpu].mpidr)); /* FALLTHROUGH */ case 3: printf(" %2d", CPU_AFF2(cpu_desc[cpu].mpidr)); /* FALLTHROUGH */ case 2: printf(" %2d", CPU_AFF1(cpu_desc[cpu].mpidr)); /* FALLTHROUGH */ case 1: case 0: /* On UP this will be zero */ printf(" %2d", CPU_AFF0(cpu_desc[cpu].mpidr)); break; } printf("\n"); /* * There is a hardware errata where, if one CPU is performing a TLB * invalidation while another is performing a store-exclusive the * store-exclusive may return the wrong status. A workaround seems * to be to use an IPI to invalidate on each CPU, however given the * limited number of affected units (pass 1.1 is the evaluation * hardware revision), and the lack of information from Cavium * this has not been implemented. * * At the time of writing this the only information is from: * https://lkml.org/lkml/2016/8/4/722 */ /* * XXX: CPU_MATCH_ERRATA_CAVIUM_THUNDER_1_1 on it's own also * triggers on pass 2.0+. */ if (cpu == 0 && CPU_VAR(PCPU_GET(midr)) == 0 && CPU_MATCH_ERRATA_CAVIUM_THUNDER_1_1) printf("WARNING: ThunderX Pass 1.1 detected.\nThis has known " "hardware bugs that may cause the incorrect operation of " "atomic operations.\n"); if (cpu != 0 && cpu_print_regs == 0) return; #define SEP_STR ((printed++) == 0) ? "" : "," /* AArch64 Instruction Set Attribute Register 0 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_ISAR0) != 0) { printed = 0; printf(" Instruction Set Attributes 0 = <"); switch (ID_AA64ISAR0_RDM(cpu_desc[cpu].id_aa64isar0)) { case ID_AA64ISAR0_RDM_NONE: break; case ID_AA64ISAR0_RDM_IMPL: printf("%sRDM", SEP_STR); break; default: printf("%sUnknown RDM", SEP_STR); } switch (ID_AA64ISAR0_ATOMIC(cpu_desc[cpu].id_aa64isar0)) { case ID_AA64ISAR0_ATOMIC_NONE: break; case ID_AA64ISAR0_ATOMIC_IMPL: printf("%sAtomic", SEP_STR); break; default: printf("%sUnknown Atomic", SEP_STR); } switch (ID_AA64ISAR0_AES(cpu_desc[cpu].id_aa64isar0)) { case ID_AA64ISAR0_AES_NONE: break; case ID_AA64ISAR0_AES_BASE: printf("%sAES", SEP_STR); break; case ID_AA64ISAR0_AES_PMULL: printf("%sAES+PMULL", SEP_STR); break; default: printf("%sUnknown AES", SEP_STR); break; } switch (ID_AA64ISAR0_SHA1(cpu_desc[cpu].id_aa64isar0)) { case ID_AA64ISAR0_SHA1_NONE: break; case ID_AA64ISAR0_SHA1_BASE: printf("%sSHA1", SEP_STR); break; default: printf("%sUnknown SHA1", SEP_STR); break; } switch (ID_AA64ISAR0_SHA2(cpu_desc[cpu].id_aa64isar0)) { case ID_AA64ISAR0_SHA2_NONE: break; case ID_AA64ISAR0_SHA2_BASE: printf("%sSHA2", SEP_STR); break; default: printf("%sUnknown SHA2", SEP_STR); break; } switch (ID_AA64ISAR0_CRC32(cpu_desc[cpu].id_aa64isar0)) { case ID_AA64ISAR0_CRC32_NONE: break; case ID_AA64ISAR0_CRC32_BASE: printf("%sCRC32", SEP_STR); break; default: printf("%sUnknown CRC32", SEP_STR); break; } if ((cpu_desc[cpu].id_aa64isar0 & ~ID_AA64ISAR0_MASK) != 0) printf("%s%#lx", SEP_STR, cpu_desc[cpu].id_aa64isar0 & ~ID_AA64ISAR0_MASK); printf(">\n"); } /* AArch64 Instruction Set Attribute Register 1 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_ISAR1) != 0) { printf(" Instruction Set Attributes 1 = <%#lx>\n", cpu_desc[cpu].id_aa64isar1); } /* AArch64 Processor Feature Register 0 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_PFR0) != 0) { printed = 0; printf(" Processor Features 0 = <"); switch (ID_AA64PFR0_GIC(cpu_desc[cpu].id_aa64pfr0)) { case ID_AA64PFR0_GIC_CPUIF_NONE: break; case ID_AA64PFR0_GIC_CPUIF_EN: printf("%sGIC", SEP_STR); break; default: printf("%sUnknown GIC interface", SEP_STR); break; } switch (ID_AA64PFR0_ADV_SIMD(cpu_desc[cpu].id_aa64pfr0)) { case ID_AA64PFR0_ADV_SIMD_NONE: break; case ID_AA64PFR0_ADV_SIMD_IMPL: printf("%sAdvSIMD", SEP_STR); break; default: printf("%sUnknown AdvSIMD", SEP_STR); break; } switch (ID_AA64PFR0_FP(cpu_desc[cpu].id_aa64pfr0)) { case ID_AA64PFR0_FP_NONE: break; case ID_AA64PFR0_FP_IMPL: printf("%sFloat", SEP_STR); break; default: printf("%sUnknown Float", SEP_STR); break; } switch (ID_AA64PFR0_EL3(cpu_desc[cpu].id_aa64pfr0)) { case ID_AA64PFR0_EL3_NONE: printf("%sNo EL3", SEP_STR); break; case ID_AA64PFR0_EL3_64: printf("%sEL3", SEP_STR); break; case ID_AA64PFR0_EL3_64_32: printf("%sEL3 32", SEP_STR); break; default: printf("%sUnknown EL3", SEP_STR); break; } switch (ID_AA64PFR0_EL2(cpu_desc[cpu].id_aa64pfr0)) { case ID_AA64PFR0_EL2_NONE: printf("%sNo EL2", SEP_STR); break; case ID_AA64PFR0_EL2_64: printf("%sEL2", SEP_STR); break; case ID_AA64PFR0_EL2_64_32: printf("%sEL2 32", SEP_STR); break; default: printf("%sUnknown EL2", SEP_STR); break; } switch (ID_AA64PFR0_EL1(cpu_desc[cpu].id_aa64pfr0)) { case ID_AA64PFR0_EL1_64: printf("%sEL1", SEP_STR); break; case ID_AA64PFR0_EL1_64_32: printf("%sEL1 32", SEP_STR); break; default: printf("%sUnknown EL1", SEP_STR); break; } switch (ID_AA64PFR0_EL0(cpu_desc[cpu].id_aa64pfr0)) { case ID_AA64PFR0_EL0_64: printf("%sEL0", SEP_STR); break; case ID_AA64PFR0_EL0_64_32: printf("%sEL0 32", SEP_STR); break; default: printf("%sUnknown EL0", SEP_STR); break; } if ((cpu_desc[cpu].id_aa64pfr0 & ~ID_AA64PFR0_MASK) != 0) printf("%s%#lx", SEP_STR, cpu_desc[cpu].id_aa64pfr0 & ~ID_AA64PFR0_MASK); printf(">\n"); } /* AArch64 Processor Feature Register 1 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_PFR1) != 0) { printf(" Processor Features 1 = <%#lx>\n", cpu_desc[cpu].id_aa64pfr1); } /* AArch64 Memory Model Feature Register 0 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_MMFR0) != 0) { printed = 0; printf(" Memory Model Features 0 = <"); switch (ID_AA64MMFR0_TGRAN4(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_TGRAN4_NONE: break; case ID_AA64MMFR0_TGRAN4_IMPL: printf("%s4k Granule", SEP_STR); break; default: printf("%sUnknown 4k Granule", SEP_STR); break; } switch (ID_AA64MMFR0_TGRAN16(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_TGRAN16_NONE: break; case ID_AA64MMFR0_TGRAN16_IMPL: printf("%s16k Granule", SEP_STR); break; default: printf("%sUnknown 16k Granule", SEP_STR); break; } switch (ID_AA64MMFR0_TGRAN64(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_TGRAN64_NONE: break; case ID_AA64MMFR0_TGRAN64_IMPL: printf("%s64k Granule", SEP_STR); break; default: printf("%sUnknown 64k Granule", SEP_STR); break; } switch (ID_AA64MMFR0_BIGEND(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_BIGEND_FIXED: break; case ID_AA64MMFR0_BIGEND_MIXED: printf("%sMixedEndian", SEP_STR); break; default: printf("%sUnknown Endian switching", SEP_STR); break; } switch (ID_AA64MMFR0_BIGEND_EL0(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_BIGEND_EL0_FIXED: break; case ID_AA64MMFR0_BIGEND_EL0_MIXED: printf("%sEL0 MixEndian", SEP_STR); break; default: printf("%sUnknown EL0 Endian switching", SEP_STR); break; } switch (ID_AA64MMFR0_S_NS_MEM(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_S_NS_MEM_NONE: break; case ID_AA64MMFR0_S_NS_MEM_DISTINCT: printf("%sS/NS Mem", SEP_STR); break; default: printf("%sUnknown S/NS Mem", SEP_STR); break; } switch (ID_AA64MMFR0_ASID_BITS(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_ASID_BITS_8: printf("%s8bit ASID", SEP_STR); break; case ID_AA64MMFR0_ASID_BITS_16: printf("%s16bit ASID", SEP_STR); break; default: printf("%sUnknown ASID", SEP_STR); break; } switch (ID_AA64MMFR0_PA_RANGE(cpu_desc[cpu].id_aa64mmfr0)) { case ID_AA64MMFR0_PA_RANGE_4G: printf("%s4GB PA", SEP_STR); break; case ID_AA64MMFR0_PA_RANGE_64G: printf("%s64GB PA", SEP_STR); break; case ID_AA64MMFR0_PA_RANGE_1T: printf("%s1TB PA", SEP_STR); break; case ID_AA64MMFR0_PA_RANGE_4T: printf("%s4TB PA", SEP_STR); break; case ID_AA64MMFR0_PA_RANGE_16T: printf("%s16TB PA", SEP_STR); break; case ID_AA64MMFR0_PA_RANGE_256T: printf("%s256TB PA", SEP_STR); break; default: printf("%sUnknown PA Range", SEP_STR); break; } if ((cpu_desc[cpu].id_aa64mmfr0 & ~ID_AA64MMFR0_MASK) != 0) printf("%s%#lx", SEP_STR, cpu_desc[cpu].id_aa64mmfr0 & ~ID_AA64MMFR0_MASK); printf(">\n"); } /* AArch64 Memory Model Feature Register 1 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_MMFR1) != 0) { printed = 0; printf(" Memory Model Features 1 = <"); switch (ID_AA64MMFR1_PAN(cpu_desc[cpu].id_aa64mmfr1)) { case ID_AA64MMFR1_PAN_NONE: break; case ID_AA64MMFR1_PAN_IMPL: printf("%sPAN", SEP_STR); break; default: printf("%sUnknown PAN", SEP_STR); break; } switch (ID_AA64MMFR1_LO(cpu_desc[cpu].id_aa64mmfr1)) { case ID_AA64MMFR1_LO_NONE: break; case ID_AA64MMFR1_LO_IMPL: printf("%sLO", SEP_STR); break; default: printf("%sUnknown LO", SEP_STR); break; } switch (ID_AA64MMFR1_HPDS(cpu_desc[cpu].id_aa64mmfr1)) { case ID_AA64MMFR1_HPDS_NONE: break; case ID_AA64MMFR1_HPDS_IMPL: printf("%sHPDS", SEP_STR); break; default: printf("%sUnknown HPDS", SEP_STR); break; } switch (ID_AA64MMFR1_VH(cpu_desc[cpu].id_aa64mmfr1)) { case ID_AA64MMFR1_VH_NONE: break; case ID_AA64MMFR1_VH_IMPL: printf("%sVHE", SEP_STR); break; default: printf("%sUnknown VHE", SEP_STR); break; } switch (ID_AA64MMFR1_VMIDBITS(cpu_desc[cpu].id_aa64mmfr1)) { case ID_AA64MMFR1_VMIDBITS_8: break; case ID_AA64MMFR1_VMIDBITS_16: printf("%s16 VMID bits", SEP_STR); break; default: printf("%sUnknown VMID bits", SEP_STR); break; } switch (ID_AA64MMFR1_HAFDBS(cpu_desc[cpu].id_aa64mmfr1)) { case ID_AA64MMFR1_HAFDBS_NONE: break; case ID_AA64MMFR1_HAFDBS_AF: printf("%sAF", SEP_STR); break; case ID_AA64MMFR1_HAFDBS_AF_DBS: printf("%sAF+DBS", SEP_STR); break; default: printf("%sUnknown Hardware update AF/DBS", SEP_STR); break; } if ((cpu_desc[cpu].id_aa64mmfr1 & ~ID_AA64MMFR1_MASK) != 0) printf("%s%#lx", SEP_STR, cpu_desc[cpu].id_aa64mmfr1 & ~ID_AA64MMFR1_MASK); printf(">\n"); } /* AArch64 Debug Feature Register 0 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_DFR0) != 0) { printed = 0; printf(" Debug Features 0 = <"); printf("%s%lu CTX Breakpoints", SEP_STR, ID_AA64DFR0_CTX_CMPS(cpu_desc[cpu].id_aa64dfr0)); printf("%s%lu Watchpoints", SEP_STR, ID_AA64DFR0_WRPS(cpu_desc[cpu].id_aa64dfr0)); printf("%s%lu Breakpoints", SEP_STR, ID_AA64DFR0_BRPS(cpu_desc[cpu].id_aa64dfr0)); switch (ID_AA64DFR0_PMU_VER(cpu_desc[cpu].id_aa64dfr0)) { case ID_AA64DFR0_PMU_VER_NONE: break; case ID_AA64DFR0_PMU_VER_3: printf("%sPMUv3", SEP_STR); break; case ID_AA64DFR0_PMU_VER_3_1: printf("%sPMUv3+16 bit evtCount", SEP_STR); break; case ID_AA64DFR0_PMU_VER_IMPL: printf("%sImplementation defined PMU", SEP_STR); break; default: printf("%sUnknown PMU", SEP_STR); break; } switch (ID_AA64DFR0_TRACE_VER(cpu_desc[cpu].id_aa64dfr0)) { case ID_AA64DFR0_TRACE_VER_NONE: break; case ID_AA64DFR0_TRACE_VER_IMPL: printf("%sTrace", SEP_STR); break; default: printf("%sUnknown Trace", SEP_STR); break; } switch (ID_AA64DFR0_DEBUG_VER(cpu_desc[cpu].id_aa64dfr0)) { case ID_AA64DFR0_DEBUG_VER_8: printf("%sDebug v8", SEP_STR); break; case ID_AA64DFR0_DEBUG_VER_8_VHE: printf("%sDebug v8+VHE", SEP_STR); break; default: printf("%sUnknown Debug", SEP_STR); break; } if (cpu_desc[cpu].id_aa64dfr0 & ~ID_AA64DFR0_MASK) printf("%s%#lx", SEP_STR, cpu_desc[cpu].id_aa64dfr0 & ~ID_AA64DFR0_MASK); printf(">\n"); } /* AArch64 Memory Model Feature Register 1 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_DFR1) != 0) { printf(" Debug Features 1 = <%#lx>\n", cpu_desc[cpu].id_aa64dfr1); } /* AArch64 Auxiliary Feature Register 0 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_AFR0) != 0) { printf(" Auxiliary Features 0 = <%#lx>\n", cpu_desc[cpu].id_aa64afr0); } /* AArch64 Auxiliary Feature Register 1 */ if (cpu == 0 || (cpu_print_regs & PRINT_ID_AA64_AFR1) != 0) { printf(" Auxiliary Features 1 = <%#lx>\n", cpu_desc[cpu].id_aa64afr1); } #undef SEP_STR } void identify_cpu(void) { u_int midr; u_int impl_id; u_int part_id; u_int cpu; size_t i; const struct cpu_parts *cpu_partsp = NULL; cpu = PCPU_GET(cpuid); midr = get_midr(); /* * Store midr to pcpu to allow fast reading * from EL0, EL1 and assembly code. */ PCPU_SET(midr, midr); impl_id = CPU_IMPL(midr); for (i = 0; i < nitems(cpu_implementers); i++) { if (impl_id == cpu_implementers[i].impl_id || cpu_implementers[i].impl_id == 0) { cpu_desc[cpu].cpu_impl = impl_id; cpu_desc[cpu].cpu_impl_name = cpu_implementers[i].impl_name; cpu_partsp = cpu_implementers[i].cpu_parts; break; } } part_id = CPU_PART(midr); for (i = 0; &cpu_partsp[i] != NULL; i++) { if (part_id == cpu_partsp[i].part_id || cpu_partsp[i].part_id == 0) { cpu_desc[cpu].cpu_part_num = part_id; cpu_desc[cpu].cpu_part_name = cpu_partsp[i].part_name; break; } } cpu_desc[cpu].cpu_revision = CPU_REV(midr); cpu_desc[cpu].cpu_variant = CPU_VAR(midr); /* Save affinity for current CPU */ cpu_desc[cpu].mpidr = get_mpidr(); CPU_AFFINITY(cpu) = cpu_desc[cpu].mpidr & CPU_AFF_MASK; cpu_desc[cpu].id_aa64dfr0 = READ_SPECIALREG(id_aa64dfr0_el1); cpu_desc[cpu].id_aa64dfr1 = READ_SPECIALREG(id_aa64dfr1_el1); cpu_desc[cpu].id_aa64isar0 = READ_SPECIALREG(id_aa64isar0_el1); cpu_desc[cpu].id_aa64isar1 = READ_SPECIALREG(id_aa64isar1_el1); cpu_desc[cpu].id_aa64mmfr0 = READ_SPECIALREG(id_aa64mmfr0_el1); cpu_desc[cpu].id_aa64mmfr1 = READ_SPECIALREG(id_aa64mmfr1_el1); cpu_desc[cpu].id_aa64pfr0 = READ_SPECIALREG(id_aa64pfr0_el1); cpu_desc[cpu].id_aa64pfr1 = READ_SPECIALREG(id_aa64pfr1_el1); if (cpu != 0) { /* * This code must run on one cpu at a time, but we are * not scheduling on the current core so implement a * simple spinlock. */ while (atomic_cmpset_acq_int(&ident_lock, 0, 1) == 0) __asm __volatile("wfe" ::: "memory"); switch (cpu_aff_levels) { case 0: if (CPU_AFF0(cpu_desc[cpu].mpidr) != CPU_AFF0(cpu_desc[0].mpidr)) cpu_aff_levels = 1; /* FALLTHROUGH */ case 1: if (CPU_AFF1(cpu_desc[cpu].mpidr) != CPU_AFF1(cpu_desc[0].mpidr)) cpu_aff_levels = 2; /* FALLTHROUGH */ case 2: if (CPU_AFF2(cpu_desc[cpu].mpidr) != CPU_AFF2(cpu_desc[0].mpidr)) cpu_aff_levels = 3; /* FALLTHROUGH */ case 3: if (CPU_AFF3(cpu_desc[cpu].mpidr) != CPU_AFF3(cpu_desc[0].mpidr)) cpu_aff_levels = 4; break; } if (cpu_desc[cpu].id_aa64afr0 != cpu_desc[0].id_aa64afr0) cpu_print_regs |= PRINT_ID_AA64_AFR0; if (cpu_desc[cpu].id_aa64afr1 != cpu_desc[0].id_aa64afr1) cpu_print_regs |= PRINT_ID_AA64_AFR1; if (cpu_desc[cpu].id_aa64dfr0 != cpu_desc[0].id_aa64dfr0) cpu_print_regs |= PRINT_ID_AA64_DFR0; if (cpu_desc[cpu].id_aa64dfr1 != cpu_desc[0].id_aa64dfr1) cpu_print_regs |= PRINT_ID_AA64_DFR1; if (cpu_desc[cpu].id_aa64isar0 != cpu_desc[0].id_aa64isar0) cpu_print_regs |= PRINT_ID_AA64_ISAR0; if (cpu_desc[cpu].id_aa64isar1 != cpu_desc[0].id_aa64isar1) cpu_print_regs |= PRINT_ID_AA64_ISAR1; if (cpu_desc[cpu].id_aa64mmfr0 != cpu_desc[0].id_aa64mmfr0) cpu_print_regs |= PRINT_ID_AA64_MMFR0; if (cpu_desc[cpu].id_aa64mmfr1 != cpu_desc[0].id_aa64mmfr1) cpu_print_regs |= PRINT_ID_AA64_MMFR1; if (cpu_desc[cpu].id_aa64pfr0 != cpu_desc[0].id_aa64pfr0) cpu_print_regs |= PRINT_ID_AA64_PFR0; if (cpu_desc[cpu].id_aa64pfr1 != cpu_desc[0].id_aa64pfr1) cpu_print_regs |= PRINT_ID_AA64_PFR1; /* Wake up the other CPUs */ atomic_store_rel_int(&ident_lock, 0); __asm __volatile("sev" ::: "memory"); } } Index: head/sys/arm64/arm64/mp_machdep.c =================================================================== --- head/sys/arm64/arm64/mp_machdep.c (revision 307209) +++ head/sys/arm64/arm64/mp_machdep.c (revision 307210) @@ -1,703 +1,697 @@ /*- * Copyright (c) 2015-2016 The FreeBSD Foundation * All rights reserved. * * This software was developed by Andrew Turner under * sponsorship from the FreeBSD Foundation. * * 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 "opt_kstack_pages.h" #include "opt_platform.h" #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef VFP #include #endif #ifdef FDT #include #include #endif #include #include "pic_if.h" typedef void intr_ipi_send_t(void *, cpuset_t, u_int); typedef void intr_ipi_handler_t(void *); #define INTR_IPI_NAMELEN (MAXCOMLEN + 1) struct intr_ipi { intr_ipi_handler_t * ii_handler; void * ii_handler_arg; intr_ipi_send_t * ii_send; void * ii_send_arg; char ii_name[INTR_IPI_NAMELEN]; u_long * ii_count; }; static struct intr_ipi ipi_sources[INTR_IPI_COUNT]; static struct intr_ipi *intr_ipi_lookup(u_int); static void intr_pic_ipi_setup(u_int, const char *, intr_ipi_handler_t *, void *); boolean_t ofw_cpu_reg(phandle_t node, u_int, cell_t *); extern struct pcpu __pcpu[]; static device_identify_t arm64_cpu_identify; static device_probe_t arm64_cpu_probe; static device_attach_t arm64_cpu_attach; static void ipi_ast(void *); static void ipi_hardclock(void *); static void ipi_preempt(void *); static void ipi_rendezvous(void *); static void ipi_stop(void *); static int ipi_handler(void *arg); struct mtx ap_boot_mtx; struct pcb stoppcbs[MAXCPU]; static device_t cpu_list[MAXCPU]; /* * Not all systems boot from the first CPU in the device tree. To work around * this we need to find which CPU we have booted from so when we later * enable the secondary CPUs we skip this one. */ static int cpu0 = -1; void mpentry(unsigned long cpuid); void init_secondary(uint64_t); uint8_t secondary_stacks[MAXCPU - 1][PAGE_SIZE * KSTACK_PAGES] __aligned(16); /* Set to 1 once we're ready to let the APs out of the pen. */ volatile int aps_ready = 0; /* Temporary variables for init_secondary() */ void *dpcpu[MAXCPU - 1]; static device_method_t arm64_cpu_methods[] = { /* Device interface */ DEVMETHOD(device_identify, arm64_cpu_identify), DEVMETHOD(device_probe, arm64_cpu_probe), DEVMETHOD(device_attach, arm64_cpu_attach), DEVMETHOD_END }; static devclass_t arm64_cpu_devclass; static driver_t arm64_cpu_driver = { "arm64_cpu", arm64_cpu_methods, 0 }; DRIVER_MODULE(arm64_cpu, cpu, arm64_cpu_driver, arm64_cpu_devclass, 0, 0); static void arm64_cpu_identify(driver_t *driver, device_t parent) { if (device_find_child(parent, "arm64_cpu", -1) != NULL) return; if (BUS_ADD_CHILD(parent, 0, "arm64_cpu", -1) == NULL) device_printf(parent, "add child failed\n"); } static int arm64_cpu_probe(device_t dev) { u_int cpuid; cpuid = device_get_unit(dev); if (cpuid >= MAXCPU || cpuid > mp_maxid) return (EINVAL); device_quiet(dev); return (0); } static int arm64_cpu_attach(device_t dev) { const uint32_t *reg; size_t reg_size; u_int cpuid; int i; cpuid = device_get_unit(dev); if (cpuid >= MAXCPU || cpuid > mp_maxid) return (EINVAL); KASSERT(cpu_list[cpuid] == NULL, ("Already have cpu %u", cpuid)); reg = cpu_get_cpuid(dev, ®_size); if (reg == NULL) return (EINVAL); if (bootverbose) { device_printf(dev, "register <"); for (i = 0; i < reg_size; i++) printf("%s%x", (i == 0) ? "" : " ", reg[i]); printf(">\n"); } /* Set the device to start it later */ cpu_list[cpuid] = dev; return (0); } static void release_aps(void *dummy __unused) { - int cpu, i; + int i; intr_pic_ipi_setup(IPI_AST, "ast", ipi_ast, NULL); intr_pic_ipi_setup(IPI_PREEMPT, "preempt", ipi_preempt, NULL); intr_pic_ipi_setup(IPI_RENDEZVOUS, "rendezvous", ipi_rendezvous, NULL); intr_pic_ipi_setup(IPI_STOP, "stop", ipi_stop, NULL); intr_pic_ipi_setup(IPI_STOP_HARD, "stop hard", ipi_stop, NULL); intr_pic_ipi_setup(IPI_HARDCLOCK, "hardclock", ipi_hardclock, NULL); atomic_store_rel_int(&aps_ready, 1); /* Wake up the other CPUs */ __asm __volatile("sev"); printf("Release APs\n"); for (i = 0; i < 2000; i++) { - if (smp_started) { - for (cpu = 0; cpu <= mp_maxid; cpu++) { - if (CPU_ABSENT(cpu)) - continue; - print_cpu_features(cpu); - } + if (smp_started) return; - } DELAY(1000); } printf("APs not started\n"); } SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL); void init_secondary(uint64_t cpu) { struct pcpu *pcpup; pcpup = &__pcpu[cpu]; /* * 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)); /* Spin until the BSP releases the APs */ while (!aps_ready) __asm __volatile("wfe"); /* Initialize curthread */ KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread")); pcpup->pc_curthread = pcpup->pc_idlethread; pcpup->pc_curpcb = pcpup->pc_idlethread->td_pcb; /* * Identify current CPU. This is necessary to setup * affinity registers and to provide support for * runtime chip identification. */ identify_cpu(); intr_pic_init_secondary(); /* Start per-CPU event timers. */ cpu_initclocks_ap(); #ifdef VFP vfp_init(); #endif dbg_monitor_init(); /* Enable interrupts */ intr_enable(); mtx_lock_spin(&ap_boot_mtx); atomic_add_rel_32(&smp_cpus, 1); if (smp_cpus == mp_ncpus) { /* enable IPI's, tlb shootdown, freezes etc */ atomic_store_rel_int(&smp_started, 1); } mtx_unlock_spin(&ap_boot_mtx); /* Enter the scheduler */ sched_throw(NULL); panic("scheduler returned us to init_secondary"); /* NOTREACHED */ } /* * Send IPI thru interrupt controller. */ static void pic_ipi_send(void *arg, cpuset_t cpus, u_int ipi) { KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__)); PIC_IPI_SEND(intr_irq_root_dev, arg, cpus, ipi); } /* * Setup IPI handler on interrupt controller. * * Not SMP coherent. */ static void intr_pic_ipi_setup(u_int ipi, const char *name, intr_ipi_handler_t *hand, void *arg) { struct intr_irqsrc *isrc; struct intr_ipi *ii; int error; KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__)); KASSERT(hand != NULL, ("%s: ipi %u no handler", __func__, ipi)); error = PIC_IPI_SETUP(intr_irq_root_dev, ipi, &isrc); if (error != 0) return; isrc->isrc_handlers++; ii = intr_ipi_lookup(ipi); KASSERT(ii->ii_count == NULL, ("%s: ipi %u reused", __func__, ipi)); ii->ii_handler = hand; ii->ii_handler_arg = arg; ii->ii_send = pic_ipi_send; ii->ii_send_arg = isrc; strlcpy(ii->ii_name, name, INTR_IPI_NAMELEN); ii->ii_count = intr_ipi_setup_counters(name); } static void intr_ipi_send(cpuset_t cpus, u_int ipi) { struct intr_ipi *ii; ii = intr_ipi_lookup(ipi); if (ii->ii_count == NULL) panic("%s: not setup IPI %u", __func__, ipi); ii->ii_send(ii->ii_send_arg, cpus, ipi); } static void ipi_ast(void *dummy __unused) { CTR0(KTR_SMP, "IPI_AST"); } static void ipi_hardclock(void *dummy __unused) { CTR1(KTR_SMP, "%s: IPI_HARDCLOCK", __func__); hardclockintr(); } static void ipi_preempt(void *dummy __unused) { CTR1(KTR_SMP, "%s: IPI_PREEMPT", __func__); sched_preempt(curthread); } static void ipi_rendezvous(void *dummy __unused) { CTR0(KTR_SMP, "IPI_RENDEZVOUS"); smp_rendezvous_action(); } static void ipi_stop(void *dummy __unused) { u_int cpu; CTR0(KTR_SMP, "IPI_STOP"); cpu = PCPU_GET(cpuid); savectx(&stoppcbs[cpu]); /* Indicate we are stopped */ CPU_SET_ATOMIC(cpu, &stopped_cpus); /* Wait for restart */ while (!CPU_ISSET(cpu, &started_cpus)) cpu_spinwait(); CPU_CLR_ATOMIC(cpu, &started_cpus); CPU_CLR_ATOMIC(cpu, &stopped_cpus); CTR0(KTR_SMP, "IPI_STOP (restart)"); } struct cpu_group * cpu_topo(void) { return (smp_topo_none()); } /* Determine if we running MP machine */ int cpu_mp_probe(void) { /* ARM64TODO: Read the u bit of mpidr_el1 to determine this */ return (1); } #ifdef FDT static boolean_t cpu_init_fdt(u_int id, phandle_t node, u_int addr_size, pcell_t *reg) { uint64_t target_cpu; struct pcpu *pcpup; vm_paddr_t pa; u_int cpuid; int err; /* Check we are able to start this cpu */ if (id > mp_maxid) return (0); KASSERT(id < MAXCPU, ("Too many CPUs")); /* We are already running on cpu 0 */ if (id == cpu0) return (1); /* * Rotate the CPU IDs to put the boot CPU as CPU 0. We keep the other * CPUs ordered as the are likely grouped into clusters so it can be * useful to keep that property, e.g. for the GICv3 driver to send * an IPI to all CPUs in the cluster. */ cpuid = id; if (cpuid < cpu0) cpuid += mp_maxid + 1; cpuid -= cpu0; pcpup = &__pcpu[cpuid]; pcpu_init(pcpup, cpuid, sizeof(struct pcpu)); dpcpu[cpuid - 1] = (void *)kmem_malloc(kernel_arena, DPCPU_SIZE, M_WAITOK | M_ZERO); dpcpu_init(dpcpu[cpuid - 1], cpuid); target_cpu = reg[0]; if (addr_size == 2) { target_cpu <<= 32; target_cpu |= reg[1]; } printf("Starting CPU %u (%lx)\n", cpuid, target_cpu); pa = pmap_extract(kernel_pmap, (vm_offset_t)mpentry); err = psci_cpu_on(target_cpu, pa, cpuid); if (err != PSCI_RETVAL_SUCCESS) { /* Panic here if INVARIANTS are enabled */ KASSERT(0, ("Failed to start CPU %u (%lx)\n", id, target_cpu)); pcpu_destroy(pcpup); kmem_free(kernel_arena, (vm_offset_t)dpcpu[cpuid - 1], DPCPU_SIZE); dpcpu[cpuid - 1] = NULL; /* Notify the user that the CPU failed to start */ printf("Failed to start CPU %u (%lx)\n", id, target_cpu); } else CPU_SET(cpuid, &all_cpus); return (1); } #endif /* Initialize and fire up non-boot processors */ void cpu_mp_start(void) { mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN); CPU_SET(0, &all_cpus); switch(arm64_bus_method) { #ifdef FDT case ARM64_BUS_FDT: KASSERT(cpu0 >= 0, ("Current CPU was not found")); ofw_cpu_early_foreach(cpu_init_fdt, true); break; #endif default: break; } } /* Introduce rest of cores to the world */ void cpu_mp_announce(void) { } static boolean_t cpu_find_cpu0_fdt(u_int id, phandle_t node, u_int addr_size, pcell_t *reg) { uint64_t mpidr_fdt, mpidr_reg; if (cpu0 < 0) { mpidr_fdt = reg[0]; if (addr_size == 2) { mpidr_fdt <<= 32; mpidr_fdt |= reg[1]; } mpidr_reg = READ_SPECIALREG(mpidr_el1); if ((mpidr_reg & 0xff00fffffful) == mpidr_fdt) cpu0 = id; } return (TRUE); } void cpu_mp_setmaxid(void) { #ifdef FDT int cores; if (arm64_bus_method == ARM64_BUS_FDT) { cores = ofw_cpu_early_foreach(cpu_find_cpu0_fdt, false); if (cores > 0) { cores = MIN(cores, MAXCPU); if (bootverbose) printf("Found %d CPUs in the device tree\n", cores); mp_ncpus = cores; mp_maxid = cores - 1; return; } } #endif if (bootverbose) printf("No CPU data, limiting to 1 core\n"); mp_ncpus = 1; mp_maxid = 0; } /* * Lookup IPI source. */ static struct intr_ipi * intr_ipi_lookup(u_int ipi) { if (ipi >= INTR_IPI_COUNT) panic("%s: no such IPI %u", __func__, ipi); return (&ipi_sources[ipi]); } /* * interrupt controller dispatch function for IPIs. It should * be called straight from the interrupt controller, when associated * interrupt source is learned. Or from anybody who has an interrupt * source mapped. */ void intr_ipi_dispatch(u_int ipi, struct trapframe *tf) { void *arg; struct intr_ipi *ii; ii = intr_ipi_lookup(ipi); if (ii->ii_count == NULL) panic("%s: not setup IPI %u", __func__, ipi); intr_ipi_increment_count(ii->ii_count, PCPU_GET(cpuid)); /* * Supply ipi filter with trapframe argument * if none is registered. */ arg = ii->ii_handler_arg != NULL ? ii->ii_handler_arg : tf; ii->ii_handler(arg); } #ifdef notyet /* * Map IPI into interrupt controller. * * Not SMP coherent. */ static int ipi_map(struct intr_irqsrc *isrc, u_int ipi) { boolean_t is_percpu; int error; if (ipi >= INTR_IPI_COUNT) panic("%s: no such IPI %u", __func__, ipi); KASSERT(intr_irq_root_dev != NULL, ("%s: no root attached", __func__)); isrc->isrc_type = INTR_ISRCT_NAMESPACE; isrc->isrc_nspc_type = INTR_IRQ_NSPC_IPI; isrc->isrc_nspc_num = ipi_next_num; error = PIC_REGISTER(intr_irq_root_dev, isrc, &is_percpu); if (error == 0) { isrc->isrc_dev = intr_irq_root_dev; ipi_next_num++; } return (error); } /* * Setup IPI handler to interrupt source. * * Note that there could be more ways how to send and receive IPIs * on a platform like fast interrupts for example. In that case, * one can call this function with ASIF_NOALLOC flag set and then * call intr_ipi_dispatch() when appropriate. * * Not SMP coherent. */ int intr_ipi_set_handler(u_int ipi, const char *name, intr_ipi_filter_t *filter, void *arg, u_int flags) { struct intr_irqsrc *isrc; int error; if (filter == NULL) return(EINVAL); isrc = intr_ipi_lookup(ipi); if (isrc->isrc_ipifilter != NULL) return (EEXIST); if ((flags & AISHF_NOALLOC) == 0) { error = ipi_map(isrc, ipi); if (error != 0) return (error); } isrc->isrc_ipifilter = filter; isrc->isrc_arg = arg; isrc->isrc_handlers = 1; isrc->isrc_count = intr_ipi_setup_counters(name); isrc->isrc_index = 0; /* it should not be used in IPI case */ if (isrc->isrc_dev != NULL) { PIC_ENABLE_INTR(isrc->isrc_dev, isrc); PIC_ENABLE_SOURCE(isrc->isrc_dev, isrc); } return (0); } #endif /* Sending IPI */ void ipi_all_but_self(u_int ipi) { cpuset_t cpus; cpus = all_cpus; CPU_CLR(PCPU_GET(cpuid), &cpus); CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); intr_ipi_send(cpus, ipi); } void ipi_cpu(int cpu, u_int ipi) { cpuset_t cpus; CPU_ZERO(&cpus); CPU_SET(cpu, &cpus); CTR3(KTR_SMP, "%s: cpu: %d, ipi: %x", __func__, cpu, ipi); intr_ipi_send(cpus, ipi); } void ipi_selected(cpuset_t cpus, u_int ipi) { CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); intr_ipi_send(cpus, ipi); }