Author: wma Date: Mon Jan 29 09:15:38 2018 New Revision: 328534 URL: https://svnweb.freebsd.org/changeset/base/328534
Log: PPC64: use hwref instead of cpuid On CHRP and PowerNV, use the interrupt server number in the cpuref and pcpu hwref field instead of the device-tree phandle and make the CPU IDs reported to the scheduler dense and with the BSP at 0. Submitted by: Wojciech Macek <[email protected]> Obtained from: Semihalf Sponsored by: IBM, QCM Technologies Differential revision: https://reviews.freebsd.org/D14011 Modified: head/sys/powerpc/powernv/platform_powernv.c head/sys/powerpc/pseries/platform_chrp.c head/sys/powerpc/pseries/xics.c Modified: head/sys/powerpc/powernv/platform_powernv.c ============================================================================== --- head/sys/powerpc/powernv/platform_powernv.c Mon Jan 29 08:10:03 2018 (r328533) +++ head/sys/powerpc/powernv/platform_powernv.c Mon Jan 29 09:15:38 2018 (r328534) @@ -1,5 +1,6 @@ /*- * Copyright (c) 2015 Nathan Whitehorn + * Copyright (c) 2017-2018 Semihalf * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -73,6 +74,7 @@ static struct cpu_group *powernv_smp_topo(platform_t p #endif static void powernv_reset(platform_t); static void powernv_cpu_idle(sbintime_t sbt); +static int powernv_cpuref_init(void); static platform_method_t powernv_methods[] = { PLATFORMMETHOD(platform_probe, powernv_probe), @@ -100,9 +102,13 @@ static platform_def_t powernv_platform = { 0 }; +static struct cpuref platform_cpuref[MAXCPU]; +static int platform_cpuref_cnt; +static int platform_cpuref_valid; + PLATFORM_DEF(powernv_platform); -static int powernv_boot_pir; +static uint64_t powernv_boot_pir; static int powernv_probe(platform_t plat) @@ -126,12 +132,20 @@ powernv_attach(platform_t plat) /* Ping OPAL again just to make sure */ opal_check(); +#if BYTE_ORDER == LITTLE_ENDIAN + opal_call(OPAL_REINIT_CPUS, 2 /* Little endian */); +#else + opal_call(OPAL_REINIT_CPUS, 1 /* Big endian */); +#endif + cpu_idle_hook = powernv_cpu_idle; powernv_boot_pir = mfspr(SPR_PIR); /* Init CPU bits */ powernv_smp_ap_init(plat); + powernv_cpuref_init(); + /* Set SLB count from device tree */ cpu = OF_peer(0); cpu = OF_child(cpu); @@ -219,147 +233,140 @@ powernv_mem_regions(platform_t plat, struct mem_region static u_long powernv_timebase_freq(platform_t plat, struct cpuref *cpuref) { - phandle_t phandle; + char buf[8]; + phandle_t cpu, dev, root; + int res; int32_t ticks = -1; - phandle = cpuref->cr_hwref; + root = OF_peer(0); + dev = OF_child(root); + while (dev != 0) { + res = OF_getprop(dev, "name", buf, sizeof(buf)); + if (res > 0 && strcmp(buf, "cpus") == 0) + break; + dev = OF_peer(dev); + } - OF_getencprop(phandle, "timebase-frequency", &ticks, sizeof(ticks)); + for (cpu = OF_child(dev); cpu != 0; cpu = OF_peer(cpu)) { + res = OF_getprop(cpu, "device_type", buf, sizeof(buf)); + if (res > 0 && strcmp(buf, "cpu") == 0) + break; + } + if (cpu == 0) + return (512000000); + OF_getencprop(cpu, "timebase-frequency", &ticks, sizeof(ticks)); + if (ticks <= 0) panic("Unable to determine timebase frequency!"); return (ticks); + } static int -powernv_smp_first_cpu(platform_t plat, struct cpuref *cpuref) +powernv_cpuref_init(void) { - char buf[8]; - phandle_t cpu, dev, root; - int res, cpuid; + phandle_t cpu, dev; + char buf[32]; + int a, res, tmp_cpuref_cnt; + static struct cpuref tmp_cpuref[MAXCPU]; + cell_t interrupt_servers[32]; + uint64_t bsp; - root = OF_peer(0); + if (platform_cpuref_valid) + return (0); - dev = OF_child(root); + dev = OF_peer(0); + dev = OF_child(dev); while (dev != 0) { res = OF_getprop(dev, "name", buf, sizeof(buf)); if (res > 0 && strcmp(buf, "cpus") == 0) break; dev = OF_peer(dev); } - if (dev == 0) { - /* - * psim doesn't have a name property on the /cpus node, - * but it can be found directly - */ - dev = OF_finddevice("/cpus"); - if (dev == 0) - return (ENOENT); - } - cpu = OF_child(dev); - - while (cpu != 0) { + bsp = 0; + tmp_cpuref_cnt = 0; + for (cpu = OF_child(dev); cpu != 0; cpu = OF_peer(cpu)) { res = OF_getprop(cpu, "device_type", buf, sizeof(buf)); - if (res > 0 && strcmp(buf, "cpu") == 0) - break; - cpu = OF_peer(cpu); - } - if (cpu == 0) - return (ENOENT); + if (res > 0 && strcmp(buf, "cpu") == 0) { + res = OF_getproplen(cpu, "ibm,ppc-interrupt-server#s"); + if (res > 0) { - cpuref->cr_hwref = cpu; - res = OF_getencprop(cpu, "ibm,ppc-interrupt-server#s", &cpuid, - sizeof(cpuid)); - if (res <= 0) - res = OF_getencprop(cpu, "reg", &cpuid, sizeof(cpuid)); - if (res <= 0) - cpuid = 0; - cpuref->cr_cpuid = cpuid; - return (0); -} + OF_getencprop(cpu, "ibm,ppc-interrupt-server#s", + interrupt_servers, res); -static int -powernv_smp_next_cpu(platform_t plat, struct cpuref *cpuref) -{ - char buf[8]; - phandle_t cpu; - int i, res, cpuid; + for (a = 0; a < res/sizeof(cell_t); a++) { + tmp_cpuref[tmp_cpuref_cnt].cr_hwref = interrupt_servers[a]; + tmp_cpuref[tmp_cpuref_cnt].cr_cpuid = tmp_cpuref_cnt; - /* Check for whether it should be the next thread */ - res = OF_getproplen(cpuref->cr_hwref, "ibm,ppc-interrupt-server#s"); - if (res > 0) { - cell_t interrupt_servers[res/sizeof(cell_t)]; - OF_getencprop(cpuref->cr_hwref, "ibm,ppc-interrupt-server#s", - interrupt_servers, res); - for (i = 0; i < res/sizeof(cell_t) - 1; i++) { - if (interrupt_servers[i] == cpuref->cr_cpuid) { - cpuref->cr_cpuid = interrupt_servers[i+1]; - return (0); + if (interrupt_servers[a] == (uint32_t)powernv_boot_pir) + bsp = tmp_cpuref_cnt; + + tmp_cpuref_cnt++; + } } } } - /* Next CPU core/package */ - cpu = OF_peer(cpuref->cr_hwref); - while (cpu != 0) { - res = OF_getprop(cpu, "device_type", buf, sizeof(buf)); - if (res > 0 && strcmp(buf, "cpu") == 0) - break; - cpu = OF_peer(cpu); + /* Map IDs, so BSP has CPUID 0 regardless of hwref */ + for (a = bsp; a < tmp_cpuref_cnt; a++) { + platform_cpuref[platform_cpuref_cnt].cr_hwref = tmp_cpuref[a].cr_hwref; + platform_cpuref[platform_cpuref_cnt].cr_cpuid = platform_cpuref_cnt; + platform_cpuref_cnt++; } - if (cpu == 0) - return (ENOENT); + for (a = 0; a < bsp; a++) { + platform_cpuref[platform_cpuref_cnt].cr_hwref = tmp_cpuref[a].cr_hwref; + platform_cpuref[platform_cpuref_cnt].cr_cpuid = platform_cpuref_cnt; + platform_cpuref_cnt++; + } - cpuref->cr_hwref = cpu; - res = OF_getencprop(cpu, "ibm,ppc-interrupt-server#s", &cpuid, - sizeof(cpuid)); - if (res <= 0) - res = OF_getencprop(cpu, "reg", &cpuid, sizeof(cpuid)); - if (res <= 0) - cpuid = 0; - cpuref->cr_cpuid = cpuid; + platform_cpuref_valid = 1; return (0); } static int -powernv_smp_get_bsp(platform_t plat, struct cpuref *cpuref) +powernv_smp_first_cpu(platform_t plat, struct cpuref *cpuref) { - phandle_t chosen; - int cpuid, res; - struct cpuref i; + if (platform_cpuref_valid == 0) + return (EINVAL); - chosen = OF_finddevice("/chosen"); - if (chosen == 0) - return (ENOENT); + cpuref->cr_cpuid = 0; + cpuref->cr_hwref = platform_cpuref[0].cr_hwref; - res = OF_getencprop(chosen, "fdtbootcpu", &cpuid, sizeof(cpuid)); - if (res < 0) - return (ENOENT); + return (0); +} - /* XXX: FDT from kexec lies sometimes. PIR seems not to. */ - if (cpuid == 0) - cpuid = powernv_boot_pir; +static int +powernv_smp_next_cpu(platform_t plat, struct cpuref *cpuref) +{ + int id; - cpuref->cr_cpuid = cpuid; + if (platform_cpuref_valid == 0) + return (EINVAL); - if (powernv_smp_first_cpu(plat, &i) != 0) + id = cpuref->cr_cpuid + 1; + if (id >= platform_cpuref_cnt) return (ENOENT); - cpuref->cr_hwref = i.cr_hwref; - do { - if (i.cr_cpuid == cpuid) { - cpuref->cr_hwref = i.cr_hwref; - break; - } - } while (powernv_smp_next_cpu(plat, &i) == 0); + cpuref->cr_cpuid = platform_cpuref[id].cr_cpuid; + cpuref->cr_hwref = platform_cpuref[id].cr_hwref; return (0); } +static int +powernv_smp_get_bsp(platform_t plat, struct cpuref *cpuref) +{ + + cpuref->cr_cpuid = platform_cpuref[0].cr_cpuid; + cpuref->cr_hwref = platform_cpuref[0].cr_hwref; + return (0); +} + #ifdef SMP static int powernv_smp_start_cpu(platform_t plat, struct pcpu *pc) @@ -369,10 +376,10 @@ powernv_smp_start_cpu(platform_t plat, struct pcpu *pc ap_pcpu = pc; powerpc_sync(); - result = opal_call(OPAL_START_CPU, pc->pc_cpuid, EXC_RST); + result = opal_call(OPAL_START_CPU, pc->pc_hwref, EXC_RST); if (result != OPAL_SUCCESS) { printf("OPAL error (%d): unable to start AP %d\n", - result, pc->pc_cpuid); + result, (int)pc->pc_hwref); return (ENXIO); } @@ -382,37 +389,49 @@ powernv_smp_start_cpu(platform_t plat, struct pcpu *pc static struct cpu_group * powernv_smp_topo(platform_t plat) { - struct pcpu *pc, *last_pc; - int i, ncores, ncpus; + char buf[8]; + phandle_t cpu, dev, root; + int res, nthreads; - ncores = ncpus = 0; - last_pc = NULL; - CPU_FOREACH(i) { - pc = pcpu_find(i); - if (pc == NULL) + root = OF_peer(0); + + dev = OF_child(root); + while (dev != 0) { + res = OF_getprop(dev, "name", buf, sizeof(buf)); + if (res > 0 && strcmp(buf, "cpus") == 0) + break; + dev = OF_peer(dev); + } + + nthreads = 1; + for (cpu = OF_child(dev); cpu != 0; cpu = OF_peer(cpu)) { + res = OF_getprop(cpu, "device_type", buf, sizeof(buf)); + if (res <= 0 || strcmp(buf, "cpu") != 0) continue; - if (last_pc == NULL || pc->pc_hwref != last_pc->pc_hwref) - ncores++; - last_pc = pc; - ncpus++; + + res = OF_getproplen(cpu, "ibm,ppc-interrupt-server#s"); + + if (res >= 0) + nthreads = res / sizeof(cell_t); + else + nthreads = 1; + break; } - if (ncpus % ncores != 0) { + if (mp_ncpus % nthreads != 0) { printf("WARNING: Irregular SMP topology. Performance may be " - "suboptimal (%d CPUS, %d cores)\n", ncpus, ncores); + "suboptimal (%d threads, %d on first core)\n", + mp_ncpus, nthreads); return (smp_topo_none()); } /* Don't do anything fancier for non-threaded SMP */ - if (ncpus == ncores) + if (nthreads == 1) return (smp_topo_none()); -#ifdef NOTYET /* smp_topo_1level() fails with non-consecutive CPU IDs */ - return (smp_topo_1level(CG_SHARE_L1, ncpus / ncores, CG_FLAG_SMT)); -#else - return (smp_topo_none()); -#endif + return (smp_topo_1level(CG_SHARE_L1, nthreads, CG_FLAG_SMT)); } + #endif static void Modified: head/sys/powerpc/pseries/platform_chrp.c ============================================================================== --- head/sys/powerpc/pseries/platform_chrp.c Mon Jan 29 08:10:03 2018 (r328533) +++ head/sys/powerpc/pseries/platform_chrp.c Mon Jan 29 09:15:38 2018 (r328534) @@ -75,6 +75,7 @@ static int chrp_smp_first_cpu(platform_t, struct cpure static int chrp_smp_next_cpu(platform_t, struct cpuref *cpuref); static int chrp_smp_get_bsp(platform_t, struct cpuref *cpuref); static void chrp_smp_ap_init(platform_t); +static int chrp_cpuref_init(void); #ifdef SMP static int chrp_smp_start_cpu(platform_t, struct pcpu *cpu); static struct cpu_group *chrp_smp_topo(platform_t plat); @@ -85,6 +86,10 @@ static void chrp_reset(platform_t); static void phyp_cpu_idle(sbintime_t sbt); #endif +static struct cpuref platform_cpuref[MAXCPU]; +static int platform_cpuref_cnt; +static int platform_cpuref_valid; + static platform_method_t chrp_methods[] = { PLATFORMMETHOD(platform_probe, chrp_probe), PLATFORMMETHOD(platform_attach, chrp_attach), @@ -159,6 +164,7 @@ chrp_attach(platform_t plat) chrp_smp_ap_init(plat); } #endif + chrp_cpuref_init(); /* Some systems (e.g. QEMU) need Open Firmware to stand down */ ofw_quiesce(); @@ -295,92 +301,30 @@ chrp_timebase_freq(platform_t plat, struct cpuref *cpu static int chrp_smp_first_cpu(platform_t plat, struct cpuref *cpuref) { - char buf[8]; - phandle_t cpu, dev, root; - int res, cpuid; - root = OF_peer(0); + if (platform_cpuref_valid == 0) + return (EINVAL); - dev = OF_child(root); - while (dev != 0) { - res = OF_getprop(dev, "name", buf, sizeof(buf)); - if (res > 0 && strcmp(buf, "cpus") == 0) - break; - dev = OF_peer(dev); - } - if (dev == 0) { - /* - * psim doesn't have a name property on the /cpus node, - * but it can be found directly - */ - dev = OF_finddevice("/cpus"); - if (dev == 0) - return (ENOENT); - } + cpuref->cr_cpuid = 0; + cpuref->cr_hwref = platform_cpuref[0].cr_hwref; - cpu = OF_child(dev); - - while (cpu != 0) { - res = OF_getprop(cpu, "device_type", buf, sizeof(buf)); - if (res > 0 && strcmp(buf, "cpu") == 0) - break; - cpu = OF_peer(cpu); - } - if (cpu == 0) - return (ENOENT); - - cpuref->cr_hwref = cpu; - res = OF_getencprop(cpu, "ibm,ppc-interrupt-server#s", &cpuid, - sizeof(cpuid)); - if (res <= 0) - res = OF_getencprop(cpu, "reg", &cpuid, sizeof(cpuid)); - if (res <= 0) - cpuid = 0; - cpuref->cr_cpuid = cpuid; - return (0); } static int chrp_smp_next_cpu(platform_t plat, struct cpuref *cpuref) { - char buf[8]; - phandle_t cpu; - int i, res, cpuid; + int id; - /* Check for whether it should be the next thread */ - res = OF_getproplen(cpuref->cr_hwref, "ibm,ppc-interrupt-server#s"); - if (res > 0) { - cell_t interrupt_servers[res/sizeof(cell_t)]; - OF_getencprop(cpuref->cr_hwref, "ibm,ppc-interrupt-server#s", - interrupt_servers, res); - for (i = 0; i < res/sizeof(cell_t) - 1; i++) { - if (interrupt_servers[i] == cpuref->cr_cpuid) { - cpuref->cr_cpuid = interrupt_servers[i+1]; - return (0); - } - } - } + if (platform_cpuref_valid == 0) + return (EINVAL); - /* Next CPU core/package */ - cpu = OF_peer(cpuref->cr_hwref); - while (cpu != 0) { - res = OF_getprop(cpu, "device_type", buf, sizeof(buf)); - if (res > 0 && strcmp(buf, "cpu") == 0) - break; - cpu = OF_peer(cpu); - } - if (cpu == 0) + id = cpuref->cr_cpuid + 1; + if (id >= platform_cpuref_cnt) return (ENOENT); - cpuref->cr_hwref = cpu; - res = OF_getencprop(cpu, "ibm,ppc-interrupt-server#s", &cpuid, - sizeof(cpuid)); - if (res <= 0) - res = OF_getencprop(cpu, "reg", &cpuid, sizeof(cpuid)); - if (res <= 0) - cpuid = 0; - cpuref->cr_cpuid = cpuid; + cpuref->cr_cpuid = platform_cpuref[id].cr_cpuid; + cpuref->cr_hwref = platform_cpuref[id].cr_hwref; return (0); } @@ -388,33 +332,60 @@ chrp_smp_next_cpu(platform_t plat, struct cpuref *cpur static int chrp_smp_get_bsp(platform_t plat, struct cpuref *cpuref) { - ihandle_t inst; - phandle_t bsp, chosen; - int res, cpuid; - chosen = OF_finddevice("/chosen"); - if (chosen == 0) - return (ENXIO); + cpuref->cr_cpuid = platform_cpuref[0].cr_cpuid; + cpuref->cr_hwref = platform_cpuref[0].cr_hwref; + return (0); +} - res = OF_getencprop(chosen, "cpu", &inst, sizeof(inst)); - if (res < 0) - return (ENXIO); +static int +chrp_cpuref_init(void) +{ + phandle_t cpu, dev; + char buf[32]; + int a, res; + cell_t interrupt_servers[32]; + uint64_t bsp; - bsp = OF_instance_to_package(inst); + if (platform_cpuref_valid) + return (0); - /* Pick the primary thread. Can it be any other? */ - cpuref->cr_hwref = bsp; - res = OF_getencprop(bsp, "ibm,ppc-interrupt-server#s", &cpuid, - sizeof(cpuid)); - if (res <= 0) - res = OF_getencprop(bsp, "reg", &cpuid, sizeof(cpuid)); - if (res <= 0) - cpuid = 0; - cpuref->cr_cpuid = cpuid; + dev = OF_peer(0); + dev = OF_child(dev); + while (dev != 0) { + res = OF_getprop(dev, "name", buf, sizeof(buf)); + if (res > 0 && strcmp(buf, "cpus") == 0) + break; + dev = OF_peer(dev); + } + bsp = 0; + for (cpu = OF_child(dev); cpu != 0; cpu = OF_peer(cpu)) { + res = OF_getprop(cpu, "device_type", buf, sizeof(buf)); + if (res > 0 && strcmp(buf, "cpu") == 0) { + res = OF_getproplen(cpu, "ibm,ppc-interrupt-server#s"); + if (res > 0) { + + + OF_getencprop(cpu, "ibm,ppc-interrupt-server#s", + interrupt_servers, res); + + for (a = 0; a < res/sizeof(cell_t); a++) { + platform_cpuref[platform_cpuref_cnt].cr_hwref = interrupt_servers[a]; + platform_cpuref[platform_cpuref_cnt].cr_cpuid = platform_cpuref_cnt; + + platform_cpuref_cnt++; + } + } + } + } + + platform_cpuref_valid = 1; + return (0); } + #ifdef SMP static int chrp_smp_start_cpu(platform_t plat, struct pcpu *pc) @@ -438,7 +409,7 @@ chrp_smp_start_cpu(platform_t plat, struct pcpu *pc) ap_pcpu = pc; powerpc_sync(); - result = rtas_call_method(start_cpu, 3, 1, pc->pc_cpuid, EXC_RST, pc, + result = rtas_call_method(start_cpu, 3, 1, pc->pc_hwref, EXC_RST, pc, &err); if (result < 0 || err != 0) { printf("RTAS error (%d/%d): unable to start AP %d\n", @@ -514,8 +485,8 @@ chrp_smp_ap_init(platform_t platform) { if (!(mfmsr() & PSL_HV)) { /* Register VPA */ - phyp_hcall(H_REGISTER_VPA, 1UL, PCPU_GET(cpuid), - splpar_vpa[PCPU_GET(cpuid)]); + phyp_hcall(H_REGISTER_VPA, 1UL, PCPU_GET(hwref), + splpar_vpa[PCPU_GET(hwref)]); /* Set interrupt priority */ phyp_hcall(H_CPPR, 0xff); Modified: head/sys/powerpc/pseries/xics.c ============================================================================== --- head/sys/powerpc/pseries/xics.c Mon Jan 29 08:10:03 2018 (r328533) +++ head/sys/powerpc/pseries/xics.c Mon Jan 29 09:15:38 2018 (r328534) @@ -263,6 +263,10 @@ xicp_bind(device_t dev, u_int irq, cpuset_t cpumask) cell_t status, cpu; int ncpus, i, error; + /* Ignore IPIs */ + if (irq == MAX_XICP_IRQS) + return; + /* * This doesn't appear to actually support affinity groups, so pick a * random CPU. @@ -281,6 +285,8 @@ xicp_bind(device_t dev, u_int irq, cpuset_t cpumask) ncpus++; } + cpu = pcpu_find(cpu)->pc_hwref; + /* XXX: super inefficient */ for (i = 0; i < sc->nintvecs; i++) { if (sc->intvecs[i].irq == irq) { @@ -312,9 +318,9 @@ xicp_dispatch(device_t dev, struct trapframe *tf) #ifdef POWERNV if (mfmsr() & PSL_HV) { - regs = xicp_mem_for_cpu(PCPU_GET(cpuid)); + regs = xicp_mem_for_cpu(PCPU_GET(hwref)); KASSERT(regs != NULL, - ("Can't find regs for CPU %d", PCPU_GET(cpuid))); + ("Can't find regs for CPU %ld", (uintptr_t)PCPU_GET(hwref))); } #endif @@ -343,7 +349,7 @@ xicp_dispatch(device_t dev, struct trapframe *tf) if (regs) bus_write_1(regs, 12, 0xff); else - phyp_hcall(H_IPI, (uint64_t)(PCPU_GET(cpuid)), + phyp_hcall(H_IPI, (uint64_t)(PCPU_GET(hwref)), 0xff); } @@ -370,7 +376,7 @@ xicp_enable(device_t dev, u_int irq, u_int vector) ("Too many XICP interrupts")); /* Bind to this CPU to start: distrib. ID is last entry in gserver# */ - cpu = PCPU_GET(cpuid); + cpu = PCPU_GET(hwref); mtx_lock(&sc->sc_mtx); sc->intvecs[sc->nintvecs].irq = irq; @@ -411,7 +417,7 @@ xicp_eoi(device_t dev, u_int irq) #ifdef POWERNV if (mfmsr() & PSL_HV) - bus_write_4(xicp_mem_for_cpu(PCPU_GET(cpuid)), 4, xirr); + bus_write_4(xicp_mem_for_cpu(PCPU_GET(hwref)), 4, xirr); else #endif phyp_hcall(H_EOI, xirr); @@ -422,6 +428,8 @@ xicp_ipi(device_t dev, u_int cpu) { #ifdef POWERNV + cpu = pcpu_find(cpu)->pc_hwref; + if (mfmsr() & PSL_HV) bus_write_1(xicp_mem_for_cpu(cpu), 12, XICP_PRIORITY); else _______________________________________________ [email protected] mailing list https://lists.freebsd.org/mailman/listinfo/svn-src-all To unsubscribe, send any mail to "[email protected]"
