diff options
Diffstat (limited to 'sys')
48 files changed, 837 insertions, 544 deletions
diff --git a/sys/amd64/acpica/acpi_wakeup.c b/sys/amd64/acpica/acpi_wakeup.c index 57341c938f14..29e66c52a2f2 100644 --- a/sys/amd64/acpica/acpi_wakeup.c +++ b/sys/amd64/acpica/acpi_wakeup.c @@ -78,7 +78,7 @@ static void acpi_stop_beep(void *); #ifdef SMP static int acpi_wakeup_ap(struct acpi_softc *, int); -static void acpi_wakeup_cpus(struct acpi_softc *, cpumask_t); +static void acpi_wakeup_cpus(struct acpi_softc *, const cpuset_t *); #endif #define WAKECODE_VADDR(sc) ((sc)->acpi_wakeaddr + (3 * PAGE_SIZE)) @@ -173,7 +173,7 @@ acpi_wakeup_ap(struct acpi_softc *sc, int cpu) #define BIOS_WARM (0x0a) static void -acpi_wakeup_cpus(struct acpi_softc *sc, cpumask_t wakeup_cpus) +acpi_wakeup_cpus(struct acpi_softc *sc, const cpuset_t *wakeup_cpus) { uint32_t mpbioswarmvec; int cpu; @@ -192,7 +192,7 @@ acpi_wakeup_cpus(struct acpi_softc *sc, cpumask_t wakeup_cpus) /* Wake up each AP. */ for (cpu = 1; cpu < mp_ncpus; cpu++) { - if ((wakeup_cpus & (1 << cpu)) == 0) + if (!CPU_ISSET(cpu, wakeup_cpus)) continue; if (acpi_wakeup_ap(sc, cpu) == 0) { /* restore the warmstart vector */ @@ -214,7 +214,7 @@ int acpi_sleep_machdep(struct acpi_softc *sc, int state) { #ifdef SMP - cpumask_t wakeup_cpus; + cpuset_t wakeup_cpus; #endif register_t cr3, rf; ACPI_STATUS status; @@ -244,10 +244,9 @@ acpi_sleep_machdep(struct acpi_softc *sc, int state) if (savectx(susppcbs[0])) { #ifdef SMP - if (wakeup_cpus != 0 && suspend_cpus(wakeup_cpus) == 0) { - device_printf(sc->acpi_dev, - "Failed to suspend APs: CPU mask = 0x%jx\n", - (uintmax_t)(wakeup_cpus & ~stopped_cpus)); + if (!CPU_EMPTY(&wakeup_cpus) && + suspend_cpus(wakeup_cpus) == 0) { + device_printf(sc->acpi_dev, "Failed to suspend APs\n"); goto out; } #endif @@ -282,8 +281,8 @@ acpi_sleep_machdep(struct acpi_softc *sc, int state) PCPU_SET(switchtime, 0); PCPU_SET(switchticks, ticks); #ifdef SMP - if (wakeup_cpus != 0) - acpi_wakeup_cpus(sc, wakeup_cpus); + if (!CPU_EMPTY(&wakeup_cpus)) + acpi_wakeup_cpus(sc, &wakeup_cpus); #endif acpi_resync_clock(sc); ret = 0; @@ -291,7 +290,7 @@ acpi_sleep_machdep(struct acpi_softc *sc, int state) out: #ifdef SMP - if (wakeup_cpus != 0) + if (!CPU_EMPTY(&wakeup_cpus)) restart_cpus(wakeup_cpus); #endif diff --git a/sys/amd64/amd64/intr_machdep.c b/sys/amd64/amd64/intr_machdep.c index 4edef81c01fa..3a8953153ff8 100644 --- a/sys/amd64/amd64/intr_machdep.c +++ b/sys/amd64/amd64/intr_machdep.c @@ -443,8 +443,7 @@ DB_SHOW_COMMAND(irqs, db_show_irqs) * allocate CPUs round-robin. */ -/* The BSP is always a valid target. */ -static cpumask_t intr_cpus = (1 << 0); +static cpuset_t intr_cpus; static int current_cpu; /* @@ -466,7 +465,7 @@ intr_next_cpu(void) current_cpu++; if (current_cpu > mp_maxid) current_cpu = 0; - } while (!(intr_cpus & (1 << current_cpu))); + } while (!CPU_ISSET(current_cpu, &intr_cpus)); mtx_unlock_spin(&icu_lock); return (apic_id); } @@ -497,7 +496,7 @@ intr_add_cpu(u_int cpu) printf("INTR: Adding local APIC %d as a target\n", cpu_apic_ids[cpu]); - intr_cpus |= (1 << cpu); + CPU_SET(cpu, &intr_cpus); } /* @@ -510,6 +509,9 @@ intr_shuffle_irqs(void *arg __unused) struct intsrc *isrc; int i; + /* The BSP is always a valid target. */ + CPU_SETOF(0, &intr_cpus); + /* Don't bother on UP. */ if (mp_ncpus == 1) return; diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c index 8e8dc67bdc1c..57e91a8444cc 100644 --- a/sys/amd64/amd64/mp_machdep.c +++ b/sys/amd64/amd64/mp_machdep.c @@ -36,6 +36,7 @@ __FBSDID("$FreeBSD$"); #include <sys/param.h> #include <sys/systm.h> #include <sys/bus.h> +#include <sys/cpuset.h> #ifdef GPROF #include <sys/gmon.h> #endif @@ -125,7 +126,7 @@ extern inthand_t IDTVEC(fast_syscall), IDTVEC(fast_syscall32); * Local data and functions. */ -static volatile cpumask_t ipi_nmi_pending; +static volatile cpuset_t ipi_nmi_pending; /* used to hold the AP's until we are ready to release them */ static struct mtx ap_boot_mtx; @@ -161,7 +162,7 @@ static void release_aps(void *dummy); static int hlt_logical_cpus; static u_int hyperthreading_cpus; /* logical cpus sharing L1 cache */ -static cpumask_t hyperthreading_cpus_mask; +static cpuset_t hyperthreading_cpus_mask; static int hyperthreading_allowed = 1; static struct sysctl_ctx_list logical_cpu_clist; static u_int bootMP_size; @@ -311,7 +312,7 @@ topo_probe(void) if (cpu_topo_probed) return; - logical_cpus_mask = 0; + CPU_ZERO(&logical_cpus_mask); if (mp_ncpus <= 1) cpu_cores = cpu_logical = 1; else if (cpu_vendor_id == CPU_VENDOR_AMD) @@ -455,7 +456,7 @@ cpu_mp_probe(void) * Always record BSP in CPU map so that the mbuf init code works * correctly. */ - all_cpus = 1; + CPU_SETOF(0, &all_cpus); if (mp_ncpus == 0) { /* * No CPUs were found, so this must be a UP system. Setup @@ -582,6 +583,7 @@ cpu_mp_announce(void) void init_secondary(void) { + cpuset_t tcpuset, tallcpus; struct pcpu *pc; struct nmi_pcpu *np; u_int64_t msr, cr0; @@ -713,19 +715,22 @@ init_secondary(void) CTR1(KTR_SMP, "SMP: AP CPU #%d Launched", PCPU_GET(cpuid)); printf("SMP: AP CPU #%d Launched!\n", PCPU_GET(cpuid)); + tcpuset = PCPU_GET(cpumask); /* Determine if we are a logical CPU. */ /* XXX Calculation depends on cpu_logical being a power of 2, e.g. 2 */ if (cpu_logical > 1 && PCPU_GET(apic_id) % cpu_logical != 0) - logical_cpus_mask |= PCPU_GET(cpumask); - + CPU_OR(&logical_cpus_mask, &tcpuset); + /* Determine if we are a hyperthread. */ if (hyperthreading_cpus > 1 && PCPU_GET(apic_id) % hyperthreading_cpus != 0) - hyperthreading_cpus_mask |= PCPU_GET(cpumask); + CPU_OR(&hyperthreading_cpus_mask, &tcpuset); /* Build our map of 'other' CPUs. */ - PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); + tallcpus = all_cpus; + CPU_NAND(&tallcpus, &tcpuset); + PCPU_SET(other_cpus, tallcpus); if (bootverbose) lapic_dump("AP"); @@ -868,6 +873,7 @@ assign_cpu_ids(void) static int start_all_aps(void) { + cpuset_t tallcpus, tcpuset; vm_offset_t va = boot_address + KERNBASE; u_int64_t *pt4, *pt3, *pt2; u_int32_t mpbioswarmvec; @@ -932,11 +938,14 @@ start_all_aps(void) panic("AP #%d (PHY# %d) failed!", cpu, apic_id); } - all_cpus |= (1 << cpu); /* record AP in CPU map */ + CPU_SET(cpu, &all_cpus); /* record AP in CPU map */ } /* build our map of 'other' CPUs */ - PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); + tallcpus = all_cpus; + tcpuset = PCPU_GET(cpumask); + CPU_NAND(&tallcpus, &tcpuset); + PCPU_SET(other_cpus, tallcpus); /* restore the warmstart vector */ *(u_int32_t *) WARMBOOT_OFF = mpbioswarmvec; @@ -1065,6 +1074,30 @@ SYSCTL_UINT(_debug_xhits, OID_AUTO, ipi_masked_range_size, CTLFLAG_RW, #endif /* COUNT_XINVLTLB_HITS */ /* + * Send an IPI to specified CPU handling the bitmap logic. + */ +static void +ipi_send_cpu(int cpu, u_int ipi) +{ + u_int bitmap, old_pending, new_pending; + + KASSERT(cpu_apic_ids[cpu] != -1, ("IPI to non-existent CPU %d", cpu)); + + if (IPI_IS_BITMAPED(ipi)) { + bitmap = 1 << ipi; + ipi = IPI_BITMAP_VECTOR; + do { + old_pending = cpu_ipi_pending[cpu]; + new_pending = old_pending | bitmap; + } while (!atomic_cmpset_int(&cpu_ipi_pending[cpu], + old_pending, new_pending)); + if (old_pending) + return; + } + lapic_ipi_vectored(ipi, cpu_apic_ids[cpu]); +} + +/* * Flush the TLB on all other CPU's */ static void @@ -1088,28 +1121,19 @@ smp_tlb_shootdown(u_int vector, vm_offset_t addr1, vm_offset_t addr2) } static void -smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2) +smp_targeted_tlb_shootdown(cpuset_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2) { - int ncpu, othercpus; + int cpu, ncpu, othercpus; othercpus = mp_ncpus - 1; - if (mask == (cpumask_t)-1) { - ncpu = othercpus; - if (ncpu < 1) + if (CPU_ISFULLSET(&mask)) { + if (othercpus < 1) return; } else { - mask &= ~PCPU_GET(cpumask); - if (mask == 0) - return; - ncpu = bitcount32(mask); - if (ncpu > othercpus) { - /* XXX this should be a panic offence */ - printf("SMP: tlb shootdown to %d other cpus (only have %d)\n", - ncpu, othercpus); - ncpu = othercpus; - } - /* XXX should be a panic, implied by mask == 0 above */ - if (ncpu < 1) + sched_pin(); + CPU_NAND(&mask, PCPU_PTR(cpumask)); + sched_unpin(); + if (CPU_EMPTY(&mask)) return; } if (!(read_rflags() & PSL_I)) @@ -1118,39 +1142,25 @@ smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_o smp_tlb_addr1 = addr1; smp_tlb_addr2 = addr2; atomic_store_rel_int(&smp_tlb_wait, 0); - if (mask == (cpumask_t)-1) + if (CPU_ISFULLSET(&mask)) { + ncpu = othercpus; ipi_all_but_self(vector); - else - ipi_selected(mask, vector); + } else { + ncpu = 0; + while ((cpu = cpusetobj_ffs(&mask)) != 0) { + cpu--; + CPU_CLR(cpu, &mask); + CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, + cpu, vector); + ipi_send_cpu(cpu, vector); + ncpu++; + } + } while (smp_tlb_wait < ncpu) ia32_pause(); mtx_unlock_spin(&smp_ipi_mtx); } -/* - * Send an IPI to specified CPU handling the bitmap logic. - */ -static void -ipi_send_cpu(int cpu, u_int ipi) -{ - u_int bitmap, old_pending, new_pending; - - KASSERT(cpu_apic_ids[cpu] != -1, ("IPI to non-existent CPU %d", cpu)); - - if (IPI_IS_BITMAPED(ipi)) { - bitmap = 1 << ipi; - ipi = IPI_BITMAP_VECTOR; - do { - old_pending = cpu_ipi_pending[cpu]; - new_pending = old_pending | bitmap; - } while (!atomic_cmpset_int(&cpu_ipi_pending[cpu], - old_pending, new_pending)); - if (old_pending) - return; - } - lapic_ipi_vectored(ipi, cpu_apic_ids[cpu]); -} - void smp_cache_flush(void) { @@ -1197,7 +1207,7 @@ smp_invlpg_range(vm_offset_t addr1, vm_offset_t addr2) } void -smp_masked_invltlb(cpumask_t mask) +smp_masked_invltlb(cpuset_t mask) { if (smp_started) { @@ -1209,7 +1219,7 @@ smp_masked_invltlb(cpumask_t mask) } void -smp_masked_invlpg(cpumask_t mask, vm_offset_t addr) +smp_masked_invlpg(cpuset_t mask, vm_offset_t addr) { if (smp_started) { @@ -1221,7 +1231,7 @@ smp_masked_invlpg(cpumask_t mask, vm_offset_t addr) } void -smp_masked_invlpg_range(cpumask_t mask, vm_offset_t addr1, vm_offset_t addr2) +smp_masked_invlpg_range(cpuset_t mask, vm_offset_t addr1, vm_offset_t addr2) { if (smp_started) { @@ -1274,7 +1284,7 @@ ipi_bitmap_handler(struct trapframe frame) * send an IPI to a set of cpus. */ void -ipi_selected(cpumask_t cpus, u_int ipi) +ipi_selected(cpuset_t cpus, u_int ipi) { int cpu; @@ -1284,12 +1294,12 @@ ipi_selected(cpumask_t cpus, u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, cpus); + CPU_OR_ATOMIC(&ipi_nmi_pending, &cpus); - CTR3(KTR_SMP, "%s: cpus: %x ipi: %x", __func__, cpus, ipi); - while ((cpu = ffs(cpus)) != 0) { + while ((cpu = cpusetobj_ffs(&cpus)) != 0) { cpu--; - cpus &= ~(1 << cpu); + CPU_CLR(cpu, &cpus); + CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi); ipi_send_cpu(cpu, ipi); } } @@ -1307,7 +1317,7 @@ ipi_cpu(int cpu, u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, 1 << cpu); + CPU_SET_ATOMIC(cpu, &ipi_nmi_pending); CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi); ipi_send_cpu(cpu, ipi); @@ -1320,8 +1330,10 @@ void ipi_all_but_self(u_int ipi) { + sched_pin(); if (IPI_IS_BITMAPED(ipi)) { ipi_selected(PCPU_GET(other_cpus), ipi); + sched_unpin(); return; } @@ -1331,7 +1343,8 @@ ipi_all_but_self(u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, PCPU_GET(other_cpus)); + CPU_OR_ATOMIC(&ipi_nmi_pending, PCPU_PTR(other_cpus)); + sched_unpin(); CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); lapic_ipi_vectored(ipi, APIC_IPI_DEST_OTHERS); @@ -1340,7 +1353,7 @@ ipi_all_but_self(u_int ipi) int ipi_nmi_handler() { - cpumask_t cpumask; + cpuset_t cpumask; /* * As long as there is not a simple way to know about a NMI's @@ -1348,11 +1361,13 @@ ipi_nmi_handler() * the global pending bitword an IPI_STOP_HARD has been issued * and should be handled. */ + sched_pin(); cpumask = PCPU_GET(cpumask); - if ((ipi_nmi_pending & cpumask) == 0) + sched_unpin(); + if (!CPU_OVERLAP(&ipi_nmi_pending, &cpumask)) return (1); - atomic_clear_int(&ipi_nmi_pending, cpumask); + CPU_NAND_ATOMIC(&ipi_nmi_pending, &cpumask); cpustop_handler(); return (0); } @@ -1364,23 +1379,25 @@ ipi_nmi_handler() void cpustop_handler(void) { - cpumask_t cpumask; + cpuset_t cpumask; u_int cpu; + sched_pin(); cpu = PCPU_GET(cpuid); cpumask = PCPU_GET(cpumask); + sched_unpin(); savectx(&stoppcbs[cpu]); /* Indicate that we are stopped */ - atomic_set_int(&stopped_cpus, cpumask); + CPU_OR_ATOMIC(&stopped_cpus, &cpumask); /* Wait for restart */ - while (!(started_cpus & cpumask)) + while (!CPU_OVERLAP(&started_cpus, &cpumask)) ia32_pause(); - atomic_clear_int(&started_cpus, cpumask); - atomic_clear_int(&stopped_cpus, cpumask); + CPU_NAND_ATOMIC(&started_cpus, &cpumask); + CPU_NAND_ATOMIC(&stopped_cpus, &cpumask); if (cpu == 0 && cpustop_restartfunc != NULL) { cpustop_restartfunc(); @@ -1395,7 +1412,7 @@ cpustop_handler(void) void cpususpend_handler(void) { - cpumask_t cpumask; + cpuset_t cpumask; register_t cr3, rf; u_int cpu; @@ -1407,7 +1424,7 @@ cpususpend_handler(void) if (savectx(susppcbs[cpu])) { wbinvd(); - atomic_set_int(&stopped_cpus, cpumask); + CPU_OR_ATOMIC(&stopped_cpus, &cpumask); } else { pmap_init_pat(); PCPU_SET(switchtime, 0); @@ -1415,11 +1432,11 @@ cpususpend_handler(void) } /* Wait for resume */ - while (!(started_cpus & cpumask)) + while (!CPU_OVERLAP(&started_cpus, &cpumask)) ia32_pause(); - atomic_clear_int(&started_cpus, cpumask); - atomic_clear_int(&stopped_cpus, cpumask); + CPU_NAND_ATOMIC(&started_cpus, &cpumask); + CPU_NAND_ATOMIC(&stopped_cpus, &cpumask); /* Restore CR3 and enable interrupts */ load_cr3(cr3); @@ -1447,30 +1464,30 @@ SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL); static int sysctl_hlt_cpus(SYSCTL_HANDLER_ARGS) { - cpumask_t mask; + cpuset_t mask; int error; mask = hlt_cpus_mask; - error = sysctl_handle_int(oidp, &mask, 0, req); + error = sysctl_handle_opaque(oidp, &mask, sizeof(mask), req); if (error || !req->newptr) return (error); - if (logical_cpus_mask != 0 && - (mask & logical_cpus_mask) == logical_cpus_mask) + if (!CPU_EMPTY(&logical_cpus_mask) && + CPU_SUBSET(&mask, &logical_cpus_mask)) hlt_logical_cpus = 1; else hlt_logical_cpus = 0; if (! hyperthreading_allowed) - mask |= hyperthreading_cpus_mask; + CPU_OR(&mask, &hyperthreading_cpus_mask); - if ((mask & all_cpus) == all_cpus) - mask &= ~(1<<0); + if (CPU_SUBSET(&mask, &all_cpus)) + CPU_CLR(0, &mask); hlt_cpus_mask = mask; return (error); } -SYSCTL_PROC(_machdep, OID_AUTO, hlt_cpus, CTLTYPE_INT|CTLFLAG_RW, - 0, 0, sysctl_hlt_cpus, "IU", +SYSCTL_PROC(_machdep, OID_AUTO, hlt_cpus, + CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, sysctl_hlt_cpus, "S", "Bitmap of CPUs to halt. 101 (binary) will halt CPUs 0 and 2."); static int @@ -1484,15 +1501,15 @@ sysctl_hlt_logical_cpus(SYSCTL_HANDLER_ARGS) return (error); if (disable) - hlt_cpus_mask |= logical_cpus_mask; + CPU_OR(&hlt_cpus_mask, &logical_cpus_mask); else - hlt_cpus_mask &= ~logical_cpus_mask; + CPU_NAND(&hlt_cpus_mask, &logical_cpus_mask); if (! hyperthreading_allowed) - hlt_cpus_mask |= hyperthreading_cpus_mask; + CPU_OR(&hlt_cpus_mask, &hyperthreading_cpus_mask); - if ((hlt_cpus_mask & all_cpus) == all_cpus) - hlt_cpus_mask &= ~(1<<0); + if (CPU_SUBSET(&hlt_cpus_mask, &all_cpus)) + CPU_CLR(0, &hlt_cpus_mask); hlt_logical_cpus = disable; return (error); @@ -1519,18 +1536,18 @@ sysctl_hyperthreading_allowed(SYSCTL_HANDLER_ARGS) #endif if (allowed) - hlt_cpus_mask &= ~hyperthreading_cpus_mask; + CPU_NAND(&hlt_cpus_mask, &hyperthreading_cpus_mask); else - hlt_cpus_mask |= hyperthreading_cpus_mask; + CPU_OR(&hlt_cpus_mask, &hyperthreading_cpus_mask); - if (logical_cpus_mask != 0 && - (hlt_cpus_mask & logical_cpus_mask) == logical_cpus_mask) + if (!CPU_EMPTY(&logical_cpus_mask) && + CPU_SUBSET(&hlt_cpus_mask, &logical_cpus_mask)) hlt_logical_cpus = 1; else hlt_logical_cpus = 0; - if ((hlt_cpus_mask & all_cpus) == all_cpus) - hlt_cpus_mask &= ~(1<<0); + if (CPU_SUBSET(&hlt_cpus_mask, &all_cpus)) + CPU_CLR(0, &hlt_cpus_mask); hyperthreading_allowed = allowed; return (error); @@ -1540,7 +1557,7 @@ static void cpu_hlt_setup(void *dummy __unused) { - if (logical_cpus_mask != 0) { + if (!CPU_EMPTY(&logical_cpus_mask)) { TUNABLE_INT_FETCH("machdep.hlt_logical_cpus", &hlt_logical_cpus); sysctl_ctx_init(&logical_cpu_clist); @@ -1554,20 +1571,21 @@ cpu_hlt_setup(void *dummy __unused) &logical_cpus_mask, 0, ""); if (hlt_logical_cpus) - hlt_cpus_mask |= logical_cpus_mask; + CPU_OR(&hlt_cpus_mask, &logical_cpus_mask); /* * If necessary for security purposes, force * hyperthreading off, regardless of the value * of hlt_logical_cpus. */ - if (hyperthreading_cpus_mask) { + if (!CPU_EMPTY(&hyperthreading_cpus_mask)) { SYSCTL_ADD_PROC(&logical_cpu_clist, SYSCTL_STATIC_CHILDREN(_machdep), OID_AUTO, "hyperthreading_allowed", CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_hyperthreading_allowed, "IU", ""); if (! hyperthreading_allowed) - hlt_cpus_mask |= hyperthreading_cpus_mask; + CPU_OR(&hlt_cpus_mask, + &hyperthreading_cpus_mask); } } } @@ -1576,7 +1594,7 @@ SYSINIT(cpu_hlt, SI_SUB_SMP, SI_ORDER_ANY, cpu_hlt_setup, NULL); int mp_grab_cpu_hlt(void) { - cpumask_t mask; + cpuset_t mask; #ifdef MP_WATCHDOG u_int cpuid; #endif @@ -1589,7 +1607,7 @@ mp_grab_cpu_hlt(void) #endif retval = 0; - while (mask & hlt_cpus_mask) { + while (CPU_OVERLAP(&mask, &hlt_cpus_mask)) { retval = 1; __asm __volatile("sti; hlt" : : : "memory"); } diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c index c9ff9bcc14ae..025ca5f981c9 100644 --- a/sys/amd64/amd64/pmap.c +++ b/sys/amd64/amd64/pmap.c @@ -123,6 +123,8 @@ __FBSDID("$FreeBSD$"); #include <sys/sysctl.h> #ifdef SMP #include <sys/smp.h> +#else +#include <sys/cpuset.h> #endif #include <vm/vm.h> @@ -581,7 +583,7 @@ pmap_bootstrap(vm_paddr_t *firstaddr) PMAP_LOCK_INIT(kernel_pmap); kernel_pmap->pm_pml4 = (pdp_entry_t *)PHYS_TO_DMAP(KPML4phys); kernel_pmap->pm_root = NULL; - kernel_pmap->pm_active = -1; /* don't allow deactivation */ + CPU_FILL(&kernel_pmap->pm_active); /* don't allow deactivation */ TAILQ_INIT(&kernel_pmap->pm_pvchunk); /* @@ -923,19 +925,20 @@ pmap_update_pde_invalidate(vm_offset_t va, pd_entry_t newpde) void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { invlpg(va); smp_invlpg(va); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) invlpg(va); - if (pmap->pm_active & other_cpus) - smp_masked_invlpg(pmap->pm_active & other_cpus, va); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invlpg(other_cpus, va); } sched_unpin(); } @@ -943,23 +946,23 @@ pmap_invalidate_page(pmap_t pmap, vm_offset_t va) void pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; vm_offset_t addr; sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); smp_invlpg_range(sva, eva); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); - if (pmap->pm_active & other_cpus) - smp_masked_invlpg_range(pmap->pm_active & other_cpus, - sva, eva); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invlpg_range(other_cpus, sva, eva); } sched_unpin(); } @@ -967,19 +970,20 @@ pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) void pmap_invalidate_all(pmap_t pmap) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { invltlb(); smp_invltlb(); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) invltlb(); - if (pmap->pm_active & other_cpus) - smp_masked_invltlb(pmap->pm_active & other_cpus); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invltlb(other_cpus); } sched_unpin(); } @@ -995,8 +999,8 @@ pmap_invalidate_cache(void) } struct pde_action { - cpumask_t store; /* processor that updates the PDE */ - cpumask_t invalidate; /* processors that invalidate their TLB */ + cpuset_t store; /* processor that updates the PDE */ + cpuset_t invalidate; /* processors that invalidate their TLB */ vm_offset_t va; pd_entry_t *pde; pd_entry_t newpde; @@ -1007,8 +1011,12 @@ pmap_update_pde_action(void *arg) { struct pde_action *act = arg; - if (act->store == PCPU_GET(cpumask)) + sched_pin(); + if (!CPU_CMP(&act->store, PCPU_PTR(cpumask))) { + sched_unpin(); pde_store(act->pde, act->newpde); + } else + sched_unpin(); } static void @@ -1016,8 +1024,12 @@ pmap_update_pde_teardown(void *arg) { struct pde_action *act = arg; - if ((act->invalidate & PCPU_GET(cpumask)) != 0) + sched_pin(); + if (CPU_OVERLAP(&act->invalidate, PCPU_PTR(cpumask))) { + sched_unpin(); pmap_update_pde_invalidate(act->va, act->newpde); + } else + sched_unpin(); } /* @@ -1032,26 +1044,28 @@ static void pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde) { struct pde_action act; - cpumask_t active, cpumask; + cpuset_t active, cpumask, other_cpus; sched_pin(); cpumask = PCPU_GET(cpumask); + other_cpus = PCPU_GET(other_cpus); if (pmap == kernel_pmap) active = all_cpus; else active = pmap->pm_active; - if ((active & PCPU_GET(other_cpus)) != 0) { + if (CPU_OVERLAP(&active, &other_cpus)) { act.store = cpumask; act.invalidate = active; act.va = va; act.pde = pde; act.newpde = newpde; - smp_rendezvous_cpus(cpumask | active, + CPU_OR(&cpumask, &active); + smp_rendezvous_cpus(cpumask, smp_no_rendevous_barrier, pmap_update_pde_action, pmap_update_pde_teardown, &act); } else { pde_store(pde, newpde); - if ((active & cpumask) != 0) + if (CPU_OVERLAP(&active, &cpumask)) pmap_update_pde_invalidate(va, newpde); } sched_unpin(); @@ -1065,7 +1079,7 @@ PMAP_INLINE void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) invlpg(va); } @@ -1074,7 +1088,7 @@ pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { vm_offset_t addr; - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); } @@ -1083,7 +1097,7 @@ PMAP_INLINE void pmap_invalidate_all(pmap_t pmap) { - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) invltlb(); } @@ -1099,7 +1113,7 @@ pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde) { pde_store(pde, newpde); - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) pmap_update_pde_invalidate(va, newpde); } #endif /* !SMP */ @@ -1607,7 +1621,7 @@ pmap_pinit0(pmap_t pmap) PMAP_LOCK_INIT(pmap); pmap->pm_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys); pmap->pm_root = NULL; - pmap->pm_active = 0; + CPU_ZERO(&pmap->pm_active); PCPU_SET(curpmap, pmap); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); @@ -1649,7 +1663,7 @@ pmap_pinit(pmap_t pmap) pmap->pm_pml4[PML4PML4I] = VM_PAGE_TO_PHYS(pml4pg) | PG_V | PG_RW | PG_A | PG_M; pmap->pm_root = NULL; - pmap->pm_active = 0; + CPU_ZERO(&pmap->pm_active); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); @@ -5087,11 +5101,11 @@ pmap_activate(struct thread *td) pmap = vmspace_pmap(td->td_proc->p_vmspace); oldpmap = PCPU_GET(curpmap); #ifdef SMP - atomic_clear_int(&oldpmap->pm_active, PCPU_GET(cpumask)); - atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask)); + CPU_NAND_ATOMIC(&oldpmap->pm_active, PCPU_PTR(cpumask)); + CPU_OR_ATOMIC(&pmap->pm_active, PCPU_PTR(cpumask)); #else - oldpmap->pm_active &= ~PCPU_GET(cpumask); - pmap->pm_active |= PCPU_GET(cpumask); + CPU_NAND(&oldpmap->pm_active, PCPU_PTR(cpumask)); + CPU_OR(&pmap->pm_active, PCPU_PTR(cpumask)); #endif cr3 = DMAP_TO_PHYS((vm_offset_t)pmap->pm_pml4); td->td_pcb->pcb_cr3 = cr3; diff --git a/sys/amd64/amd64/vm_machdep.c b/sys/amd64/amd64/vm_machdep.c index 972484a80a08..13f5cd06b2b3 100644 --- a/sys/amd64/amd64/vm_machdep.c +++ b/sys/amd64/amd64/vm_machdep.c @@ -59,6 +59,7 @@ __FBSDID("$FreeBSD$"); #include <sys/mutex.h> #include <sys/pioctl.h> #include <sys/proc.h> +#include <sys/sched.h> #include <sys/sf_buf.h> #include <sys/smp.h> #include <sys/sysctl.h> @@ -70,6 +71,7 @@ __FBSDID("$FreeBSD$"); #include <machine/cpu.h> #include <machine/md_var.h> #include <machine/pcb.h> +#include <machine/smp.h> #include <machine/specialreg.h> #include <machine/tss.h> @@ -512,11 +514,13 @@ cpu_set_user_tls(struct thread *td, void *tls_base) static void cpu_reset_proxy() { + cpuset_t tcrp; cpu_reset_proxy_active = 1; while (cpu_reset_proxy_active == 1) ; /* Wait for other cpu to see that we've started */ - stop_cpus((1<<cpu_reset_proxyid)); + CPU_SETOF(cpu_reset_proxyid, &tcrp); + stop_cpus(tcrp); printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); DELAY(1000000); cpu_reset_real(); @@ -527,24 +531,28 @@ void cpu_reset() { #ifdef SMP - cpumask_t map; + cpuset_t map; u_int cnt; if (smp_active) { - map = PCPU_GET(other_cpus) & ~stopped_cpus; - if (map != 0) { + sched_pin(); + map = PCPU_GET(other_cpus); + CPU_NAND(&map, &stopped_cpus); + if (!CPU_EMPTY(&map)) { printf("cpu_reset: Stopping other CPUs\n"); stop_cpus(map); } if (PCPU_GET(cpuid) != 0) { cpu_reset_proxyid = PCPU_GET(cpuid); + sched_unpin(); cpustop_restartfunc = cpu_reset_proxy; cpu_reset_proxy_active = 0; printf("cpu_reset: Restarting BSP\n"); /* Restart CPU #0. */ - atomic_store_rel_int(&started_cpus, 1 << 0); + CPU_SETOF(0, &started_cpus); + wmb(); cnt = 0; while (cpu_reset_proxy_active == 0 && cnt < 10000000) @@ -556,7 +564,8 @@ cpu_reset() while (1); /* NOTREACHED */ - } + } else + sched_unpin(); DELAY(1000000); } diff --git a/sys/amd64/include/_types.h b/sys/amd64/include/_types.h index 89d2e861b418..13dc3ea28a7a 100644 --- a/sys/amd64/include/_types.h +++ b/sys/amd64/include/_types.h @@ -61,7 +61,6 @@ typedef unsigned long __uint64_t; * Standard type definitions. */ typedef __int32_t __clock_t; /* clock()... */ -typedef unsigned int __cpumask_t; typedef __int64_t __critical_t; typedef double __double_t; typedef float __float_t; diff --git a/sys/amd64/include/param.h b/sys/amd64/include/param.h index 9a742f94e9b9..92689bae21ab 100644 --- a/sys/amd64/include/param.h +++ b/sys/amd64/include/param.h @@ -64,7 +64,7 @@ #endif #if defined(SMP) || defined(KLD_MODULE) -#define MAXCPU 32 +#define MAXCPU 64 #else #define MAXCPU 1 #endif diff --git a/sys/amd64/include/pmap.h b/sys/amd64/include/pmap.h index 7a628513af65..1b8108a3e3c6 100644 --- a/sys/amd64/include/pmap.h +++ b/sys/amd64/include/pmap.h @@ -152,6 +152,7 @@ #ifndef LOCORE #include <sys/queue.h> +#include <sys/_cpuset.h> #include <sys/_lock.h> #include <sys/_mutex.h> @@ -251,7 +252,7 @@ struct pmap { struct mtx pm_mtx; pml4_entry_t *pm_pml4; /* KVA of level 4 page table */ TAILQ_HEAD(,pv_chunk) pm_pvchunk; /* list of mappings in pmap */ - cpumask_t pm_active; /* active on cpus */ + cpuset_t pm_active; /* active on cpus */ /* spare u_int here due to padding */ struct pmap_statistics pm_stats; /* pmap statistics */ vm_page_t pm_root; /* spare page table pages */ diff --git a/sys/amd64/include/smp.h b/sys/amd64/include/smp.h index ec107f931055..a009b35b150e 100644 --- a/sys/amd64/include/smp.h +++ b/sys/amd64/include/smp.h @@ -63,17 +63,17 @@ void ipi_all_but_self(u_int ipi); void ipi_bitmap_handler(struct trapframe frame); void ipi_cpu(int cpu, u_int ipi); int ipi_nmi_handler(void); -void ipi_selected(cpumask_t cpus, u_int ipi); +void ipi_selected(cpuset_t cpus, u_int ipi); u_int mp_bootaddress(u_int); int mp_grab_cpu_hlt(void); void smp_cache_flush(void); void smp_invlpg(vm_offset_t addr); -void smp_masked_invlpg(cpumask_t mask, vm_offset_t addr); +void smp_masked_invlpg(cpuset_t mask, vm_offset_t addr); void smp_invlpg_range(vm_offset_t startva, vm_offset_t endva); -void smp_masked_invlpg_range(cpumask_t mask, vm_offset_t startva, +void smp_masked_invlpg_range(cpuset_t mask, vm_offset_t startva, vm_offset_t endva); void smp_invltlb(void); -void smp_masked_invltlb(cpumask_t mask); +void smp_masked_invltlb(cpuset_t mask); #endif /* !LOCORE */ #endif /* SMP */ diff --git a/sys/arm/arm/pmap.c b/sys/arm/arm/pmap.c index f33a692e7302..16487b08ef9d 100644 --- a/sys/arm/arm/pmap.c +++ b/sys/arm/arm/pmap.c @@ -2395,7 +2395,7 @@ pmap_bootstrap(vm_offset_t firstaddr, vm_offset_t lastaddr, struct pv_addr *l1pt cpu_cpwait(); PMAP_LOCK_INIT(kernel_pmap); - kernel_pmap->pm_active = -1; + CPU_FILL(&kernel_pmap->pm_active); kernel_pmap->pm_domain = PMAP_DOMAIN_KERNEL; TAILQ_INIT(&kernel_pmap->pm_pvlist); @@ -3826,7 +3826,7 @@ pmap_pinit(pmap_t pmap) pmap_alloc_l1(pmap); bzero(pmap->pm_l2, sizeof(pmap->pm_l2)); - pmap->pm_active = 0; + CPU_ZERO(&pmap->pm_active); TAILQ_INIT(&pmap->pm_pvlist); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); diff --git a/sys/arm/include/_types.h b/sys/arm/include/_types.h index 48dd2a784fd7..d8386f3225d0 100644 --- a/sys/arm/include/_types.h +++ b/sys/arm/include/_types.h @@ -67,7 +67,6 @@ typedef unsigned long long __uint64_t; * Standard type definitions. */ typedef __uint32_t __clock_t; /* clock()... */ -typedef unsigned int __cpumask_t; typedef __int32_t __critical_t; typedef double __double_t; typedef double __float_t; diff --git a/sys/arm/include/pmap.h b/sys/arm/include/pmap.h index 701390a33852..3d63432e332a 100644 --- a/sys/arm/include/pmap.h +++ b/sys/arm/include/pmap.h @@ -62,6 +62,7 @@ #ifndef LOCORE #include <sys/queue.h> +#include <sys/_cpuset.h> #include <sys/_lock.h> #include <sys/_mutex.h> @@ -134,7 +135,7 @@ struct pmap { struct l1_ttable *pm_l1; struct l2_dtable *pm_l2[L2_SIZE]; pd_entry_t *pm_pdir; /* KVA of page directory */ - cpumask_t pm_active; /* active on cpus */ + cpuset_t pm_active; /* active on cpus */ struct pmap_statistics pm_stats; /* pmap statictics */ TAILQ_HEAD(,pv_entry) pm_pvlist; /* list of mappings in pmap */ }; diff --git a/sys/cddl/dev/cyclic/i386/cyclic_machdep.c b/sys/cddl/dev/cyclic/i386/cyclic_machdep.c index 1c18fc14e96c..dc6cb712f228 100644 --- a/sys/cddl/dev/cyclic/i386/cyclic_machdep.c +++ b/sys/cddl/dev/cyclic/i386/cyclic_machdep.c @@ -121,7 +121,9 @@ static void reprogram(cyb_arg_t arg, hrtime_t exp) static void xcall(cyb_arg_t arg, cpu_t *c, cyc_func_t func, void *param) { + cpuset_t cpus; - smp_rendezvous_cpus((cpumask_t) (1 << c->cpuid), + CPU_SETOF(c->cpuid, &cpus); + smp_rendezvous_cpus(cpus, smp_no_rendevous_barrier, func, smp_no_rendevous_barrier, param); } diff --git a/sys/cddl/dev/dtrace/amd64/dtrace_subr.c b/sys/cddl/dev/dtrace/amd64/dtrace_subr.c index 102ea0495aff..bc4b8b62b985 100644 --- a/sys/cddl/dev/dtrace/amd64/dtrace_subr.c +++ b/sys/cddl/dev/dtrace/amd64/dtrace_subr.c @@ -113,12 +113,12 @@ dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) void dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) { - cpumask_t cpus; + cpuset_t cpus; if (cpu == DTRACE_CPUALL) cpus = all_cpus; else - cpus = (cpumask_t)1 << cpu; + CPU_SETOF(cpu, &cpus); smp_rendezvous_cpus(cpus, smp_no_rendevous_barrier, func, smp_no_rendevous_barrier, arg); @@ -394,7 +394,7 @@ dtrace_gethrtime_init(void *arg) { struct pcpu *pc; uint64_t tsc_f; - cpumask_t map; + cpuset_t map; int i; /* @@ -432,7 +432,8 @@ dtrace_gethrtime_init(void *arg) continue; pc = pcpu_find(i); - map = PCPU_GET(cpumask) | pc->pc_cpumask; + map = PCPU_GET(cpumask); + CPU_OR(&map, &pc->pc_cpumask); smp_rendezvous_cpus(map, dtrace_gethrtime_init_sync, dtrace_gethrtime_init_cpu, diff --git a/sys/cddl/dev/dtrace/i386/dtrace_subr.c b/sys/cddl/dev/dtrace/i386/dtrace_subr.c index 50793c4da4fc..7d460bc2e03b 100644 --- a/sys/cddl/dev/dtrace/i386/dtrace_subr.c +++ b/sys/cddl/dev/dtrace/i386/dtrace_subr.c @@ -30,6 +30,7 @@ #include <sys/param.h> #include <sys/systm.h> #include <sys/types.h> +#include <sys/cpuset.h> #include <sys/kernel.h> #include <sys/malloc.h> #include <sys/kmem.h> @@ -113,12 +114,12 @@ dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) void dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) { - cpumask_t cpus; + cpuset_t cpus; if (cpu == DTRACE_CPUALL) cpus = all_cpus; else - cpus = (cpumask_t)1 << cpu; + CPU_SETOF(cpu, &cpus); smp_rendezvous_cpus(cpus, smp_no_rendevous_barrier, func, smp_no_rendevous_barrier, arg); @@ -392,9 +393,9 @@ dtrace_gethrtime_init_cpu(void *arg) static void dtrace_gethrtime_init(void *arg) { + cpuset_t map; struct pcpu *pc; uint64_t tsc_f; - cpumask_t map; int i; /* @@ -432,7 +433,8 @@ dtrace_gethrtime_init(void *arg) continue; pc = pcpu_find(i); - map = PCPU_GET(cpumask) | pc->pc_cpumask; + map = PCPU_GET(cpumask); + CPU_OR(&map, &pc->pc_cpumask); smp_rendezvous_cpus(map, dtrace_gethrtime_init_sync, dtrace_gethrtime_init_cpu, diff --git a/sys/dev/hwpmc/hwpmc_mod.c b/sys/dev/hwpmc/hwpmc_mod.c index d6225d8f5ae3..4cfcea87fa5b 100644 --- a/sys/dev/hwpmc/hwpmc_mod.c +++ b/sys/dev/hwpmc/hwpmc_mod.c @@ -1991,7 +1991,7 @@ pmc_hook_handler(struct thread *td, int function, void *arg) * had already processed the interrupt). We don't * lose the interrupt sample. */ - atomic_clear_int(&pmc_cpumask, (1 << PCPU_GET(cpuid))); + CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmc_cpumask); pmc_process_samples(PCPU_GET(cpuid)); break; @@ -4083,7 +4083,7 @@ pmc_process_interrupt(int cpu, struct pmc *pm, struct trapframe *tf, done: /* mark CPU as needing processing */ - atomic_set_int(&pmc_cpumask, (1 << cpu)); + CPU_SET_ATOMIC(cpu, &pmc_cpumask); return (error); } @@ -4193,7 +4193,7 @@ pmc_process_samples(int cpu) break; if (ps->ps_nsamples == PMC_SAMPLE_INUSE) { /* Need a rescan at a later time. */ - atomic_set_int(&pmc_cpumask, (1 << cpu)); + CPU_SET_ATOMIC(cpu, &pmc_cpumask); break; } @@ -4782,7 +4782,7 @@ pmc_cleanup(void) PMCDBG(MOD,INI,0, "%s", "cleanup"); /* switch off sampling */ - pmc_cpumask = 0; + CPU_ZERO(&pmc_cpumask); pmc_intr = NULL; sx_xlock(&pmc_sx); diff --git a/sys/dev/xen/control/control.c b/sys/dev/xen/control/control.c index c03d5365530b..0f4418142684 100644 --- a/sys/dev/xen/control/control.c +++ b/sys/dev/xen/control/control.c @@ -203,24 +203,29 @@ xctrl_suspend() unsigned long max_pfn, start_info_mfn; #ifdef SMP - cpumask_t map; + struct thread *td; + cpuset_t map; /* * Bind us to CPU 0 and stop any other VCPUs. */ - thread_lock(curthread); - sched_bind(curthread, 0); - thread_unlock(curthread); + td = curthread; + thread_lock(td); + sched_bind(td, 0); + thread_unlock(td); KASSERT(PCPU_GET(cpuid) == 0, ("xen_suspend: not running on cpu 0")); - map = PCPU_GET(other_cpus) & ~stopped_cpus; - if (map) + sched_pin(); + map = PCPU_GET(other_cpus); + sched_unpin(); + CPU_NAND(&map, &stopped_cpus); + if (!CPU_EMPTY(&map)) stop_cpus(map); #endif if (DEVICE_SUSPEND(root_bus) != 0) { printf("xen_suspend: device_suspend failed\n"); #ifdef SMP - if (map) + if (!CPU_EMPTY(&map)) restart_cpus(map); #endif return; @@ -289,7 +294,7 @@ xctrl_suspend() thread_lock(curthread); sched_unbind(curthread); thread_unlock(curthread); - if (map) + if (!CPU_EMPTY(&map)) restart_cpus(map); #endif } diff --git a/sys/geom/eli/g_eli.c b/sys/geom/eli/g_eli.c index 8b22ee58d72b..6466aef72ab7 100644 --- a/sys/geom/eli/g_eli.c +++ b/sys/geom/eli/g_eli.c @@ -661,7 +661,7 @@ static int g_eli_cpu_is_disabled(int cpu) { #ifdef SMP - return ((hlt_cpus_mask & (1 << cpu)) != 0); + return (CPU_ISSET(cpu, &hlt_cpus_mask)); #else return (0); #endif diff --git a/sys/i386/i386/intr_machdep.c b/sys/i386/i386/intr_machdep.c index 77b80048620f..56529f7b5b03 100644 --- a/sys/i386/i386/intr_machdep.c +++ b/sys/i386/i386/intr_machdep.c @@ -409,8 +409,7 @@ DB_SHOW_COMMAND(irqs, db_show_irqs) * allocate CPUs round-robin. */ -/* The BSP is always a valid target. */ -static cpumask_t intr_cpus = (1 << 0); +static cpuset_t intr_cpus; static int current_cpu; /* @@ -432,7 +431,7 @@ intr_next_cpu(void) current_cpu++; if (current_cpu > mp_maxid) current_cpu = 0; - } while (!(intr_cpus & (1 << current_cpu))); + } while (!CPU_ISSET(current_cpu, &intr_cpus)); mtx_unlock_spin(&icu_lock); return (apic_id); } @@ -463,7 +462,7 @@ intr_add_cpu(u_int cpu) printf("INTR: Adding local APIC %d as a target\n", cpu_apic_ids[cpu]); - intr_cpus |= (1 << cpu); + CPU_SET(cpu, &intr_cpus); } /* @@ -483,6 +482,9 @@ intr_shuffle_irqs(void *arg __unused) return; #endif + /* The BSP is always a valid target. */ + CPU_SETOF(0, &intr_cpus); + /* Don't bother on UP. */ if (mp_ncpus == 1) return; diff --git a/sys/i386/i386/mp_machdep.c b/sys/i386/i386/mp_machdep.c index a07b06c53f75..dfe6434518f0 100644 --- a/sys/i386/i386/mp_machdep.c +++ b/sys/i386/i386/mp_machdep.c @@ -51,6 +51,7 @@ __FBSDID("$FreeBSD$"); #include <sys/systm.h> #include <sys/bus.h> #include <sys/cons.h> /* cngetc() */ +#include <sys/cpuset.h> #ifdef GPROF #include <sys/gmon.h> #endif @@ -172,7 +173,7 @@ static u_long *ipi_hardclock_counts[MAXCPU]; * Local data and functions. */ -static volatile cpumask_t ipi_nmi_pending; +static volatile cpuset_t ipi_nmi_pending; /* used to hold the AP's until we are ready to release them */ static struct mtx ap_boot_mtx; @@ -209,7 +210,7 @@ static void release_aps(void *dummy); static int hlt_logical_cpus; static u_int hyperthreading_cpus; /* logical cpus sharing L1 cache */ -static cpumask_t hyperthreading_cpus_mask; +static cpuset_t hyperthreading_cpus_mask; static int hyperthreading_allowed = 1; static struct sysctl_ctx_list logical_cpu_clist; @@ -358,7 +359,7 @@ topo_probe(void) if (cpu_topo_probed) return; - logical_cpus_mask = 0; + CPU_ZERO(&logical_cpus_mask); if (mp_ncpus <= 1) cpu_cores = cpu_logical = 1; else if (cpu_vendor_id == CPU_VENDOR_AMD) @@ -500,7 +501,7 @@ cpu_mp_probe(void) * Always record BSP in CPU map so that the mbuf init code works * correctly. */ - all_cpus = 1; + CPU_SETOF(0, &all_cpus); if (mp_ncpus == 0) { /* * No CPUs were found, so this must be a UP system. Setup @@ -631,6 +632,7 @@ cpu_mp_announce(void) void init_secondary(void) { + cpuset_t tcpuset, tallcpus; struct pcpu *pc; vm_offset_t addr; int gsel_tss; @@ -755,19 +757,22 @@ init_secondary(void) CTR1(KTR_SMP, "SMP: AP CPU #%d Launched", PCPU_GET(cpuid)); printf("SMP: AP CPU #%d Launched!\n", PCPU_GET(cpuid)); + tcpuset = PCPU_GET(cpumask); /* Determine if we are a logical CPU. */ /* XXX Calculation depends on cpu_logical being a power of 2, e.g. 2 */ if (cpu_logical > 1 && PCPU_GET(apic_id) % cpu_logical != 0) - logical_cpus_mask |= PCPU_GET(cpumask); + CPU_OR(&logical_cpus_mask, &tcpuset); /* Determine if we are a hyperthread. */ if (hyperthreading_cpus > 1 && PCPU_GET(apic_id) % hyperthreading_cpus != 0) - hyperthreading_cpus_mask |= PCPU_GET(cpumask); + CPU_OR(&hyperthreading_cpus_mask, &tcpuset); /* Build our map of 'other' CPUs. */ - PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); + tallcpus = all_cpus; + CPU_NAND(&tallcpus, &tcpuset); + PCPU_SET(other_cpus, tallcpus); if (bootverbose) lapic_dump("AP"); @@ -904,6 +909,7 @@ assign_cpu_ids(void) static int start_all_aps(void) { + cpuset_t tallcpus; #ifndef PC98 u_char mpbiosreason; #endif @@ -963,11 +969,13 @@ start_all_aps(void) } CHECK_PRINT("trace"); /* show checkpoints */ - all_cpus |= (1 << cpu); /* record AP in CPU map */ + CPU_SET(cpu, &all_cpus); /* record AP in CPU map */ } /* build our map of 'other' CPUs */ - PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); + tallcpus = all_cpus; + CPU_NAND(&tallcpus, PCPU_PTR(cpumask)); + PCPU_SET(other_cpus, tallcpus); /* restore the warmstart vector */ *(u_int32_t *) WARMBOOT_OFF = mpbioswarmvec; @@ -1164,6 +1172,30 @@ SYSCTL_INT(_debug_xhits, OID_AUTO, ipi_masked_range_size, CTLFLAG_RW, #endif /* COUNT_XINVLTLB_HITS */ /* + * Send an IPI to specified CPU handling the bitmap logic. + */ +static void +ipi_send_cpu(int cpu, u_int ipi) +{ + u_int bitmap, old_pending, new_pending; + + KASSERT(cpu_apic_ids[cpu] != -1, ("IPI to non-existent CPU %d", cpu)); + + if (IPI_IS_BITMAPED(ipi)) { + bitmap = 1 << ipi; + ipi = IPI_BITMAP_VECTOR; + do { + old_pending = cpu_ipi_pending[cpu]; + new_pending = old_pending | bitmap; + } while (!atomic_cmpset_int(&cpu_ipi_pending[cpu], + old_pending, new_pending)); + if (old_pending) + return; + } + lapic_ipi_vectored(ipi, cpu_apic_ids[cpu]); +} + +/* * Flush the TLB on all other CPU's */ static void @@ -1187,28 +1219,19 @@ smp_tlb_shootdown(u_int vector, vm_offset_t addr1, vm_offset_t addr2) } static void -smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2) +smp_targeted_tlb_shootdown(cpuset_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2) { - int ncpu, othercpus; + int cpu, ncpu, othercpus; othercpus = mp_ncpus - 1; - if (mask == (u_int)-1) { - ncpu = othercpus; - if (ncpu < 1) + if (CPU_ISFULLSET(&mask)) { + if (othercpus < 1) return; } else { - mask &= ~PCPU_GET(cpumask); - if (mask == 0) - return; - ncpu = bitcount32(mask); - if (ncpu > othercpus) { - /* XXX this should be a panic offence */ - printf("SMP: tlb shootdown to %d other cpus (only have %d)\n", - ncpu, othercpus); - ncpu = othercpus; - } - /* XXX should be a panic, implied by mask == 0 above */ - if (ncpu < 1) + sched_pin(); + CPU_NAND(&mask, PCPU_PTR(cpumask)); + sched_unpin(); + if (CPU_EMPTY(&mask)) return; } if (!(read_eflags() & PSL_I)) @@ -1217,39 +1240,25 @@ smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_o smp_tlb_addr1 = addr1; smp_tlb_addr2 = addr2; atomic_store_rel_int(&smp_tlb_wait, 0); - if (mask == (u_int)-1) + if (CPU_ISFULLSET(&mask)) { + ncpu = othercpus; ipi_all_but_self(vector); - else - ipi_selected(mask, vector); + } else { + ncpu = 0; + while ((cpu = cpusetobj_ffs(&mask)) != 0) { + cpu--; + CPU_CLR(cpu, &mask); + CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, + vector); + ipi_send_cpu(cpu, vector); + ncpu++; + } + } while (smp_tlb_wait < ncpu) ia32_pause(); mtx_unlock_spin(&smp_ipi_mtx); } -/* - * Send an IPI to specified CPU handling the bitmap logic. - */ -static void -ipi_send_cpu(int cpu, u_int ipi) -{ - u_int bitmap, old_pending, new_pending; - - KASSERT(cpu_apic_ids[cpu] != -1, ("IPI to non-existent CPU %d", cpu)); - - if (IPI_IS_BITMAPED(ipi)) { - bitmap = 1 << ipi; - ipi = IPI_BITMAP_VECTOR; - do { - old_pending = cpu_ipi_pending[cpu]; - new_pending = old_pending | bitmap; - } while (!atomic_cmpset_int(&cpu_ipi_pending[cpu], - old_pending, new_pending)); - if (old_pending) - return; - } - lapic_ipi_vectored(ipi, cpu_apic_ids[cpu]); -} - void smp_cache_flush(void) { @@ -1296,7 +1305,7 @@ smp_invlpg_range(vm_offset_t addr1, vm_offset_t addr2) } void -smp_masked_invltlb(cpumask_t mask) +smp_masked_invltlb(cpuset_t mask) { if (smp_started) { @@ -1308,7 +1317,7 @@ smp_masked_invltlb(cpumask_t mask) } void -smp_masked_invlpg(cpumask_t mask, vm_offset_t addr) +smp_masked_invlpg(cpuset_t mask, vm_offset_t addr) { if (smp_started) { @@ -1320,7 +1329,7 @@ smp_masked_invlpg(cpumask_t mask, vm_offset_t addr) } void -smp_masked_invlpg_range(cpumask_t mask, vm_offset_t addr1, vm_offset_t addr2) +smp_masked_invlpg_range(cpuset_t mask, vm_offset_t addr1, vm_offset_t addr2) { if (smp_started) { @@ -1373,7 +1382,7 @@ ipi_bitmap_handler(struct trapframe frame) * send an IPI to a set of cpus. */ void -ipi_selected(cpumask_t cpus, u_int ipi) +ipi_selected(cpuset_t cpus, u_int ipi) { int cpu; @@ -1383,12 +1392,12 @@ ipi_selected(cpumask_t cpus, u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, cpus); + CPU_OR_ATOMIC(&ipi_nmi_pending, &cpus); - CTR3(KTR_SMP, "%s: cpus: %x ipi: %x", __func__, cpus, ipi); - while ((cpu = ffs(cpus)) != 0) { + while ((cpu = cpusetobj_ffs(&cpus)) != 0) { cpu--; - cpus &= ~(1 << cpu); + CPU_CLR(cpu, &cpus); + CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi); ipi_send_cpu(cpu, ipi); } } @@ -1406,7 +1415,7 @@ ipi_cpu(int cpu, u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, 1 << cpu); + CPU_SET_ATOMIC(cpu, &ipi_nmi_pending); CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi); ipi_send_cpu(cpu, ipi); @@ -1419,8 +1428,10 @@ void ipi_all_but_self(u_int ipi) { + sched_pin(); if (IPI_IS_BITMAPED(ipi)) { ipi_selected(PCPU_GET(other_cpus), ipi); + sched_unpin(); return; } @@ -1430,7 +1441,9 @@ ipi_all_but_self(u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, PCPU_GET(other_cpus)); + CPU_OR_ATOMIC(&ipi_nmi_pending, PCPU_PTR(other_cpus)); + sched_unpin(); + CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); lapic_ipi_vectored(ipi, APIC_IPI_DEST_OTHERS); } @@ -1438,7 +1451,7 @@ ipi_all_but_self(u_int ipi) int ipi_nmi_handler() { - cpumask_t cpumask; + cpuset_t cpumask; /* * As long as there is not a simple way to know about a NMI's @@ -1446,11 +1459,13 @@ ipi_nmi_handler() * the global pending bitword an IPI_STOP_HARD has been issued * and should be handled. */ + sched_pin(); cpumask = PCPU_GET(cpumask); - if ((ipi_nmi_pending & cpumask) == 0) + sched_unpin(); + if (!CPU_OVERLAP(&ipi_nmi_pending, &cpumask)) return (1); - atomic_clear_int(&ipi_nmi_pending, cpumask); + CPU_NAND_ATOMIC(&ipi_nmi_pending, &cpumask); cpustop_handler(); return (0); } @@ -1462,23 +1477,25 @@ ipi_nmi_handler() void cpustop_handler(void) { - cpumask_t cpumask; + cpuset_t cpumask; u_int cpu; + sched_pin(); cpu = PCPU_GET(cpuid); cpumask = PCPU_GET(cpumask); + sched_unpin(); savectx(&stoppcbs[cpu]); /* Indicate that we are stopped */ - atomic_set_int(&stopped_cpus, cpumask); + CPU_OR_ATOMIC(&stopped_cpus, &cpumask); /* Wait for restart */ - while (!(started_cpus & cpumask)) + while (!CPU_OVERLAP(&started_cpus, &cpumask)) ia32_pause(); - atomic_clear_int(&started_cpus, cpumask); - atomic_clear_int(&stopped_cpus, cpumask); + CPU_NAND_ATOMIC(&started_cpus, &cpumask); + CPU_NAND_ATOMIC(&stopped_cpus, &cpumask); if (cpu == 0 && cpustop_restartfunc != NULL) { cpustop_restartfunc(); @@ -1505,30 +1522,30 @@ SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL); static int sysctl_hlt_cpus(SYSCTL_HANDLER_ARGS) { - cpumask_t mask; + cpuset_t mask; int error; mask = hlt_cpus_mask; - error = sysctl_handle_int(oidp, &mask, 0, req); + error = sysctl_handle_opaque(oidp, &mask, sizeof(mask), req); if (error || !req->newptr) return (error); - if (logical_cpus_mask != 0 && - (mask & logical_cpus_mask) == logical_cpus_mask) + if (!CPU_EMPTY(&logical_cpus_mask) && + CPU_SUBSET(&mask, &logical_cpus_mask)) hlt_logical_cpus = 1; else hlt_logical_cpus = 0; if (! hyperthreading_allowed) - mask |= hyperthreading_cpus_mask; + CPU_OR(&mask, &hyperthreading_cpus_mask); - if ((mask & all_cpus) == all_cpus) - mask &= ~(1<<0); + if (CPU_SUBSET(&mask, &all_cpus)) + CPU_CLR(0, &mask); hlt_cpus_mask = mask; return (error); } -SYSCTL_PROC(_machdep, OID_AUTO, hlt_cpus, CTLTYPE_INT|CTLFLAG_RW, - 0, 0, sysctl_hlt_cpus, "IU", +SYSCTL_PROC(_machdep, OID_AUTO, hlt_cpus, + CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0, sysctl_hlt_cpus, "S", "Bitmap of CPUs to halt. 101 (binary) will halt CPUs 0 and 2."); static int @@ -1542,15 +1559,15 @@ sysctl_hlt_logical_cpus(SYSCTL_HANDLER_ARGS) return (error); if (disable) - hlt_cpus_mask |= logical_cpus_mask; + CPU_OR(&hlt_cpus_mask, &logical_cpus_mask); else - hlt_cpus_mask &= ~logical_cpus_mask; + CPU_NAND(&hlt_cpus_mask, &logical_cpus_mask); if (! hyperthreading_allowed) - hlt_cpus_mask |= hyperthreading_cpus_mask; + CPU_OR(&hlt_cpus_mask, &hyperthreading_cpus_mask); - if ((hlt_cpus_mask & all_cpus) == all_cpus) - hlt_cpus_mask &= ~(1<<0); + if (CPU_SUBSET(&hlt_cpus_mask, &all_cpus)) + CPU_CLR(0, &hlt_cpus_mask); hlt_logical_cpus = disable; return (error); @@ -1577,18 +1594,18 @@ sysctl_hyperthreading_allowed(SYSCTL_HANDLER_ARGS) #endif if (allowed) - hlt_cpus_mask &= ~hyperthreading_cpus_mask; + CPU_NAND(&hlt_cpus_mask, &hyperthreading_cpus_mask); else - hlt_cpus_mask |= hyperthreading_cpus_mask; + CPU_OR(&hlt_cpus_mask, &hyperthreading_cpus_mask); - if (logical_cpus_mask != 0 && - (hlt_cpus_mask & logical_cpus_mask) == logical_cpus_mask) + if (!CPU_EMPTY(&logical_cpus_mask) && + CPU_SUBSET(&hlt_cpus_mask, &logical_cpus_mask)) hlt_logical_cpus = 1; else hlt_logical_cpus = 0; - if ((hlt_cpus_mask & all_cpus) == all_cpus) - hlt_cpus_mask &= ~(1<<0); + if (CPU_SUBSET(&hlt_cpus_mask, &all_cpus)) + CPU_CLR(0, &hlt_cpus_mask); hyperthreading_allowed = allowed; return (error); @@ -1598,7 +1615,7 @@ static void cpu_hlt_setup(void *dummy __unused) { - if (logical_cpus_mask != 0) { + if (!CPU_EMPTY(&logical_cpus_mask)) { TUNABLE_INT_FETCH("machdep.hlt_logical_cpus", &hlt_logical_cpus); sysctl_ctx_init(&logical_cpu_clist); @@ -1612,20 +1629,21 @@ cpu_hlt_setup(void *dummy __unused) &logical_cpus_mask, 0, ""); if (hlt_logical_cpus) - hlt_cpus_mask |= logical_cpus_mask; + CPU_OR(&hlt_cpus_mask, &logical_cpus_mask); /* * If necessary for security purposes, force * hyperthreading off, regardless of the value * of hlt_logical_cpus. */ - if (hyperthreading_cpus_mask) { + if (!CPU_EMPTY(&hyperthreading_cpus_mask)) { SYSCTL_ADD_PROC(&logical_cpu_clist, SYSCTL_STATIC_CHILDREN(_machdep), OID_AUTO, "hyperthreading_allowed", CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_hyperthreading_allowed, "IU", ""); if (! hyperthreading_allowed) - hlt_cpus_mask |= hyperthreading_cpus_mask; + CPU_OR(&hlt_cpus_mask, + &hyperthreading_cpus_mask); } } } @@ -1634,7 +1652,7 @@ SYSINIT(cpu_hlt, SI_SUB_SMP, SI_ORDER_ANY, cpu_hlt_setup, NULL); int mp_grab_cpu_hlt(void) { - cpumask_t mask; + cpuset_t mask; #ifdef MP_WATCHDOG u_int cpuid; #endif @@ -1647,7 +1665,7 @@ mp_grab_cpu_hlt(void) #endif retval = 0; - while (mask & hlt_cpus_mask) { + while (CPU_OVERLAP(&mask, &hlt_cpus_mask)) { retval = 1; __asm __volatile("sti; hlt" : : : "memory"); } diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c index e1fe137e4d31..c85f1387030e 100644 --- a/sys/i386/i386/pmap.c +++ b/sys/i386/i386/pmap.c @@ -125,6 +125,8 @@ __FBSDID("$FreeBSD$"); #include <sys/sysctl.h> #ifdef SMP #include <sys/smp.h> +#else +#include <sys/cpuset.h> #endif #include <vm/vm.h> @@ -386,7 +388,7 @@ pmap_bootstrap(vm_paddr_t firstaddr) kernel_pmap->pm_pdpt = (pdpt_entry_t *) (KERNBASE + (u_int)IdlePDPT); #endif kernel_pmap->pm_root = NULL; - kernel_pmap->pm_active = -1; /* don't allow deactivation */ + CPU_FILL(&kernel_pmap->pm_active); /* don't allow deactivation */ TAILQ_INIT(&kernel_pmap->pm_pvchunk); LIST_INIT(&allpmaps); @@ -930,19 +932,20 @@ pmap_update_pde_invalidate(vm_offset_t va, pd_entry_t newpde) void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { invlpg(va); smp_invlpg(va); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) invlpg(va); - if (pmap->pm_active & other_cpus) - smp_masked_invlpg(pmap->pm_active & other_cpus, va); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invlpg(other_cpus, va); } sched_unpin(); } @@ -950,23 +953,23 @@ pmap_invalidate_page(pmap_t pmap, vm_offset_t va) void pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; vm_offset_t addr; sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); smp_invlpg_range(sva, eva); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); - if (pmap->pm_active & other_cpus) - smp_masked_invlpg_range(pmap->pm_active & other_cpus, - sva, eva); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invlpg_range(other_cpus, sva, eva); } sched_unpin(); } @@ -974,19 +977,20 @@ pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) void pmap_invalidate_all(pmap_t pmap) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { invltlb(); smp_invltlb(); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) invltlb(); - if (pmap->pm_active & other_cpus) - smp_masked_invltlb(pmap->pm_active & other_cpus); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invltlb(other_cpus); } sched_unpin(); } @@ -1002,8 +1006,8 @@ pmap_invalidate_cache(void) } struct pde_action { - cpumask_t store; /* processor that updates the PDE */ - cpumask_t invalidate; /* processors that invalidate their TLB */ + cpuset_t store; /* processor that updates the PDE */ + cpuset_t invalidate; /* processors that invalidate their TLB */ vm_offset_t va; pd_entry_t *pde; pd_entry_t newpde; @@ -1016,7 +1020,10 @@ pmap_update_pde_kernel(void *arg) pd_entry_t *pde; pmap_t pmap; - if (act->store == PCPU_GET(cpumask)) + sched_pin(); + if (!CPU_CMP(&act->store, PCPU_PTR(cpumask))) { + sched_unpin(); + /* * Elsewhere, this operation requires allpmaps_lock for * synchronization. Here, it does not because it is being @@ -1026,6 +1033,8 @@ pmap_update_pde_kernel(void *arg) pde = pmap_pde(pmap, act->va); pde_store(pde, act->newpde); } + } else + sched_unpin(); } static void @@ -1033,8 +1042,12 @@ pmap_update_pde_user(void *arg) { struct pde_action *act = arg; - if (act->store == PCPU_GET(cpumask)) + sched_pin(); + if (!CPU_CMP(&act->store, PCPU_PTR(cpumask))) { + sched_unpin(); pde_store(act->pde, act->newpde); + } else + sched_unpin(); } static void @@ -1042,8 +1055,12 @@ pmap_update_pde_teardown(void *arg) { struct pde_action *act = arg; - if ((act->invalidate & PCPU_GET(cpumask)) != 0) + sched_pin(); + if (CPU_OVERLAP(&act->invalidate, PCPU_PTR(cpumask))) { + sched_unpin(); pmap_update_pde_invalidate(act->va, act->newpde); + } else + sched_unpin(); } /* @@ -1058,21 +1075,23 @@ static void pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde) { struct pde_action act; - cpumask_t active, cpumask; + cpuset_t active, cpumask, other_cpus; sched_pin(); cpumask = PCPU_GET(cpumask); + other_cpus = PCPU_GET(other_cpus); if (pmap == kernel_pmap) active = all_cpus; else active = pmap->pm_active; - if ((active & PCPU_GET(other_cpus)) != 0) { + if (CPU_OVERLAP(&active, &other_cpus)) { act.store = cpumask; act.invalidate = active; act.va = va; act.pde = pde; act.newpde = newpde; - smp_rendezvous_cpus(cpumask | active, + CPU_OR(&cpumask, &active); + smp_rendezvous_cpus(cpumask, smp_no_rendevous_barrier, pmap == kernel_pmap ? pmap_update_pde_kernel : pmap_update_pde_user, pmap_update_pde_teardown, &act); @@ -1081,7 +1100,7 @@ pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde) pmap_kenter_pde(va, newpde); else pde_store(pde, newpde); - if ((active & cpumask) != 0) + if (CPU_OVERLAP(&active, &cpumask)) pmap_update_pde_invalidate(va, newpde); } sched_unpin(); @@ -1095,7 +1114,7 @@ PMAP_INLINE void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) invlpg(va); } @@ -1104,7 +1123,7 @@ pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { vm_offset_t addr; - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); } @@ -1113,7 +1132,7 @@ PMAP_INLINE void pmap_invalidate_all(pmap_t pmap) { - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) invltlb(); } @@ -1132,7 +1151,7 @@ pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde) pmap_kenter_pde(va, newpde); else pde_store(pde, newpde); - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) pmap_update_pde_invalidate(va, newpde); } #endif /* !SMP */ @@ -1689,7 +1708,7 @@ pmap_pinit0(pmap_t pmap) pmap->pm_pdpt = (pdpt_entry_t *)(KERNBASE + (vm_offset_t)IdlePDPT); #endif pmap->pm_root = NULL; - pmap->pm_active = 0; + CPU_ZERO(&pmap->pm_active); PCPU_SET(curpmap, pmap); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); @@ -1770,7 +1789,7 @@ pmap_pinit(pmap_t pmap) #endif } - pmap->pm_active = 0; + CPU_ZERO(&pmap->pm_active); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); @@ -4985,11 +5004,11 @@ pmap_activate(struct thread *td) pmap = vmspace_pmap(td->td_proc->p_vmspace); oldpmap = PCPU_GET(curpmap); #if defined(SMP) - atomic_clear_int(&oldpmap->pm_active, PCPU_GET(cpumask)); - atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask)); + CPU_NAND_ATOMIC(&oldpmap->pm_active, PCPU_PTR(cpumask)); + CPU_OR_ATOMIC(&pmap->pm_active, PCPU_PTR(cpumask)); #else - oldpmap->pm_active &= ~1; - pmap->pm_active |= 1; + CPU_NAND(&oldpmap->pm_active, PCPU_PTR(cpumask)); + CPU_OR(&pmap->pm_active, PCPU_PTR(cpumask)); #endif #ifdef PAE cr3 = vtophys(pmap->pm_pdpt); diff --git a/sys/i386/i386/vm_machdep.c b/sys/i386/i386/vm_machdep.c index 232e1a1f308e..a084e09526aa 100644 --- a/sys/i386/i386/vm_machdep.c +++ b/sys/i386/i386/vm_machdep.c @@ -573,11 +573,13 @@ kvtop(void *addr) static void cpu_reset_proxy() { + cpuset_t tcrp; cpu_reset_proxy_active = 1; while (cpu_reset_proxy_active == 1) ; /* Wait for other cpu to see that we've started */ - stop_cpus((1<<cpu_reset_proxyid)); + CPU_SETOF(cpu_reset_proxyid, &tcrp); + stop_cpus(tcrp); printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); DELAY(1000000); cpu_reset_real(); @@ -596,25 +598,29 @@ cpu_reset() #endif #ifdef SMP - cpumask_t map; + cpuset_t map; u_int cnt; if (smp_active) { - map = PCPU_GET(other_cpus) & ~stopped_cpus; - if (map != 0) { + sched_pin(); + map = PCPU_GET(other_cpus); + CPU_NAND(&map, &stopped_cpus); + if (!CPU_EMPTY(&map)) { printf("cpu_reset: Stopping other CPUs\n"); stop_cpus(map); } if (PCPU_GET(cpuid) != 0) { cpu_reset_proxyid = PCPU_GET(cpuid); + sched_unpin(); cpustop_restartfunc = cpu_reset_proxy; cpu_reset_proxy_active = 0; printf("cpu_reset: Restarting BSP\n"); /* Restart CPU #0. */ /* XXX: restart_cpus(1 << 0); */ - atomic_store_rel_int(&started_cpus, (1 << 0)); + CPU_SETOF(0, &started_cpus); + wmb(); cnt = 0; while (cpu_reset_proxy_active == 0 && cnt < 10000000) @@ -626,7 +632,8 @@ cpu_reset() while (1); /* NOTREACHED */ - } + } else + sched_unpin(); DELAY(1000000); } @@ -795,7 +802,7 @@ sf_buf_alloc(struct vm_page *m, int flags) struct sf_head *hash_list; struct sf_buf *sf; #ifdef SMP - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; #endif int error; @@ -867,22 +874,23 @@ sf_buf_alloc(struct vm_page *m, int flags) */ #ifdef SMP if ((opte & (PG_V | PG_A)) == (PG_V | PG_A)) - sf->cpumask = 0; + CPU_ZERO(&sf->cpumask); shootdown: sched_pin(); cpumask = PCPU_GET(cpumask); - if ((sf->cpumask & cpumask) == 0) { - sf->cpumask |= cpumask; + if (!CPU_OVERLAP(&cpumask, &sf->cpumask)) { + CPU_OR(&sf->cpumask, &cpumask); invlpg(sf->kva); } if ((flags & SFB_CPUPRIVATE) == 0) { - other_cpus = PCPU_GET(other_cpus) & ~sf->cpumask; - if (other_cpus != 0) { - sf->cpumask |= other_cpus; + other_cpus = PCPU_GET(other_cpus); + CPU_NAND(&other_cpus, &sf->cpumask); + if (!CPU_EMPTY(&other_cpus)) { + CPU_OR(&sf->cpumask, &other_cpus); smp_masked_invlpg(other_cpus, sf->kva); } } - sched_unpin(); + sched_unpin(); #else if ((opte & (PG_V | PG_A)) == (PG_V | PG_A)) pmap_invalidate_page(kernel_pmap, sf->kva); diff --git a/sys/i386/include/_types.h b/sys/i386/include/_types.h index 7a969fedf160..3194fd691f67 100644 --- a/sys/i386/include/_types.h +++ b/sys/i386/include/_types.h @@ -69,7 +69,6 @@ typedef unsigned long long __uint64_t; * Standard type definitions. */ typedef unsigned long __clock_t; /* clock()... */ -typedef unsigned int __cpumask_t; typedef __int32_t __critical_t; typedef long double __double_t; typedef long double __float_t; diff --git a/sys/i386/include/pmap.h b/sys/i386/include/pmap.h index eeada2e400f4..3012a000ae5c 100644 --- a/sys/i386/include/pmap.h +++ b/sys/i386/include/pmap.h @@ -155,6 +155,7 @@ #ifndef LOCORE #include <sys/queue.h> +#include <sys/_cpuset.h> #include <sys/_lock.h> #include <sys/_mutex.h> @@ -433,7 +434,7 @@ struct pmap { struct mtx pm_mtx; pd_entry_t *pm_pdir; /* KVA of page directory */ TAILQ_HEAD(,pv_chunk) pm_pvchunk; /* list of mappings in pmap */ - cpumask_t pm_active; /* active on cpus */ + cpuset_t pm_active; /* active on cpus */ struct pmap_statistics pm_stats; /* pmap statistics */ LIST_ENTRY(pmap) pm_list; /* List of all pmaps */ #ifdef PAE diff --git a/sys/i386/include/sf_buf.h b/sys/i386/include/sf_buf.h index 7bc1095c9590..415dcbb86e63 100644 --- a/sys/i386/include/sf_buf.h +++ b/sys/i386/include/sf_buf.h @@ -29,6 +29,7 @@ #ifndef _MACHINE_SF_BUF_H_ #define _MACHINE_SF_BUF_H_ +#include <sys/_cpuset.h> #include <sys/queue.h> struct vm_page; @@ -40,7 +41,7 @@ struct sf_buf { vm_offset_t kva; /* va of mapping */ int ref_count; /* usage of this mapping */ #ifdef SMP - cpumask_t cpumask; /* cpus on which mapping is valid */ + cpuset_t cpumask; /* cpus on which mapping is valid */ #endif }; diff --git a/sys/i386/include/smp.h b/sys/i386/include/smp.h index b512e00b4228..cdec9c87bace 100644 --- a/sys/i386/include/smp.h +++ b/sys/i386/include/smp.h @@ -64,17 +64,17 @@ void ipi_bitmap_handler(struct trapframe frame); #endif void ipi_cpu(int cpu, u_int ipi); int ipi_nmi_handler(void); -void ipi_selected(cpumask_t cpus, u_int ipi); +void ipi_selected(cpuset_t cpus, u_int ipi); u_int mp_bootaddress(u_int); int mp_grab_cpu_hlt(void); void smp_cache_flush(void); void smp_invlpg(vm_offset_t addr); -void smp_masked_invlpg(cpumask_t mask, vm_offset_t addr); +void smp_masked_invlpg(cpuset_t mask, vm_offset_t addr); void smp_invlpg_range(vm_offset_t startva, vm_offset_t endva); -void smp_masked_invlpg_range(cpumask_t mask, vm_offset_t startva, +void smp_masked_invlpg_range(cpuset_t mask, vm_offset_t startva, vm_offset_t endva); void smp_invltlb(void); -void smp_masked_invltlb(cpumask_t mask); +void smp_masked_invltlb(cpuset_t mask); #ifdef XEN void ipi_to_irq_init(void); diff --git a/sys/i386/xen/mp_machdep.c b/sys/i386/xen/mp_machdep.c index 670d110af49c..1146ac2351e6 100644 --- a/sys/i386/xen/mp_machdep.c +++ b/sys/i386/xen/mp_machdep.c @@ -52,6 +52,7 @@ __FBSDID("$FreeBSD$"); #include <sys/systm.h> #include <sys/bus.h> #include <sys/cons.h> /* cngetc() */ +#include <sys/cpuset.h> #ifdef GPROF #include <sys/gmon.h> #endif @@ -116,7 +117,7 @@ volatile int smp_tlb_wait; typedef void call_data_func_t(uintptr_t , uintptr_t); static u_int logical_cpus; -static volatile cpumask_t ipi_nmi_pending; +static volatile cpuset_t ipi_nmi_pending; /* used to hold the AP's until we are ready to release them */ static struct mtx ap_boot_mtx; @@ -149,7 +150,7 @@ static int start_ap(int apic_id); static void release_aps(void *dummy); static u_int hyperthreading_cpus; -static cpumask_t hyperthreading_cpus_mask; +static cpuset_t hyperthreading_cpus_mask; extern void Xhypervisor_callback(void); extern void failsafe_callback(void); @@ -238,7 +239,7 @@ cpu_mp_probe(void) * Always record BSP in CPU map so that the mbuf init code works * correctly. */ - all_cpus = 1; + CPU_SETOF(0, &all_cpus); if (mp_ncpus == 0) { /* * No CPUs were found, so this must be a UP system. Setup @@ -292,7 +293,8 @@ cpu_mp_start(void) start_all_aps(); /* Setup the initial logical CPUs info. */ - logical_cpus = logical_cpus_mask = 0; + logical_cpus = 0; + CPU_ZERO(&logical_cpus_mask); if (cpu_feature & CPUID_HTT) logical_cpus = (cpu_procinfo & CPUID_HTT_CORES) >> 16; @@ -512,6 +514,7 @@ xen_smp_intr_init_cpus(void *unused) void init_secondary(void) { + cpuset_t tcpuset, tallcpus; vm_offset_t addr; int gsel_tss; @@ -591,18 +594,21 @@ init_secondary(void) CTR1(KTR_SMP, "SMP: AP CPU #%d Launched", PCPU_GET(cpuid)); printf("SMP: AP CPU #%d Launched!\n", PCPU_GET(cpuid)); + tcpuset = PCPU_GET(cpumask); /* Determine if we are a logical CPU. */ if (logical_cpus > 1 && PCPU_GET(apic_id) % logical_cpus != 0) - logical_cpus_mask |= PCPU_GET(cpumask); + CPU_OR(&logical_cpus_mask, &tcpuset); /* Determine if we are a hyperthread. */ if (hyperthreading_cpus > 1 && PCPU_GET(apic_id) % hyperthreading_cpus != 0) - hyperthreading_cpus_mask |= PCPU_GET(cpumask); + CPU_OR(&hyperthreading_cpus_mask, &tcpuset); /* Build our map of 'other' CPUs. */ - PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); + tallcpus = all_cpus; + CPU_NAND(&tallcpus, &tcpuset); + PCPU_SET(other_cpus, tallcpus); #if 0 if (bootverbose) lapic_dump("AP"); @@ -713,6 +719,7 @@ assign_cpu_ids(void) int start_all_aps(void) { + cpuset_t tallcpus; int x,apic_id, cpu; struct pcpu *pc; @@ -766,12 +773,14 @@ start_all_aps(void) panic("bye-bye"); } - all_cpus |= (1 << cpu); /* record AP in CPU map */ + CPU_SET(cpu, &all_cpus); /* record AP in CPU map */ } /* build our map of 'other' CPUs */ - PCPU_SET(other_cpus, all_cpus & ~PCPU_GET(cpumask)); + tallcpus = all_cpus; + CPU_NAND(&tallcpus, PCPU_PTR(cpumask)); + PCPU_SET(other_cpus, tallcpus); pmap_invalidate_range(kernel_pmap, 0, NKPT * NBPDR - 1); @@ -1000,29 +1009,20 @@ smp_tlb_shootdown(u_int vector, vm_offset_t addr1, vm_offset_t addr2) } static void -smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2) +smp_targeted_tlb_shootdown(cpuset_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2) { - int ncpu, othercpus; + int cpu, ncpu, othercpus; struct _call_data data; othercpus = mp_ncpus - 1; - if (mask == (u_int)-1) { - ncpu = othercpus; - if (ncpu < 1) + if (CPU_ISFULLSET(&mask)) { + if (othercpus < 1) return; } else { - mask &= ~PCPU_GET(cpumask); - if (mask == 0) - return; - ncpu = bitcount32(mask); - if (ncpu > othercpus) { - /* XXX this should be a panic offence */ - printf("SMP: tlb shootdown to %d other cpus (only have %d)\n", - ncpu, othercpus); - ncpu = othercpus; - } - /* XXX should be a panic, implied by mask == 0 above */ - if (ncpu < 1) + critical_enter(); + CPU_NAND(&mask, PCPU_PTR(cpumask)); + critical_exit(); + if (CPU_EMPTY(&mask)) return; } if (!(read_eflags() & PSL_I)) @@ -1034,10 +1034,20 @@ smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_o call_data->arg1 = addr1; call_data->arg2 = addr2; atomic_store_rel_int(&smp_tlb_wait, 0); - if (mask == (u_int)-1) + if (CPU_ISFULLSET(&mask)) { + ncpu = othercpus; ipi_all_but_self(vector); - else - ipi_selected(mask, vector); + } else { + ncpu = 0; + while ((cpu = cpusetobj_ffs(&mask)) != 0) { + cpu--; + CPU_CLR(cpu, &mask); + CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, + vector); + ipi_send_cpu(cpu, vector); + ncpu++; + } + } while (smp_tlb_wait < ncpu) ia32_pause(); call_data = NULL; @@ -1080,7 +1090,7 @@ smp_invlpg_range(vm_offset_t addr1, vm_offset_t addr2) } void -smp_masked_invltlb(cpumask_t mask) +smp_masked_invltlb(cpuset_t mask) { if (smp_started) { @@ -1089,7 +1099,7 @@ smp_masked_invltlb(cpumask_t mask) } void -smp_masked_invlpg(cpumask_t mask, vm_offset_t addr) +smp_masked_invlpg(cpuset_t mask, vm_offset_t addr) { if (smp_started) { @@ -1098,7 +1108,7 @@ smp_masked_invlpg(cpumask_t mask, vm_offset_t addr) } void -smp_masked_invlpg_range(cpumask_t mask, vm_offset_t addr1, vm_offset_t addr2) +smp_masked_invlpg_range(cpuset_t mask, vm_offset_t addr1, vm_offset_t addr2) { if (smp_started) { @@ -1110,7 +1120,7 @@ smp_masked_invlpg_range(cpumask_t mask, vm_offset_t addr1, vm_offset_t addr2) * send an IPI to a set of cpus. */ void -ipi_selected(cpumask_t cpus, u_int ipi) +ipi_selected(cpuset_t cpus, u_int ipi) { int cpu; @@ -1120,11 +1130,11 @@ ipi_selected(cpumask_t cpus, u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, cpus); + CPU_OR_ATOMIC(&ipi_nmi_pending, &cpus); - while ((cpu = ffs(cpus)) != 0) { + while ((cpu = cpusetobj_ffs(&cpus)) != 0) { cpu--; - cpus &= ~(1 << cpu); + CPU_CLR(cpu, &cpus); CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi); ipi_send_cpu(cpu, ipi); } @@ -1143,7 +1153,7 @@ ipi_cpu(int cpu, u_int ipi) * Set the mask of receiving CPUs for this purpose. */ if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, 1 << cpu); + CPU_SET_ATOMIC(cpu, &ipi_nmi_pending); CTR3(KTR_SMP, "%s: cpu: %d ipi: %x", __func__, cpu, ipi); ipi_send_cpu(cpu, ipi); @@ -1155,23 +1165,27 @@ ipi_cpu(int cpu, u_int ipi) void ipi_all_but_self(u_int ipi) { + cpuset_t other_cpus; /* * IPI_STOP_HARD maps to a NMI and the trap handler needs a bit * of help in order to understand what is the source. * Set the mask of receiving CPUs for this purpose. */ + sched_pin(); + other_cpus = PCPU_GET(other_cpus); + sched_unpin(); if (ipi == IPI_STOP_HARD) - atomic_set_int(&ipi_nmi_pending, PCPU_GET(other_cpus)); + CPU_OR_ATOMIC(&ipi_nmi_pending, &other_cpus); CTR2(KTR_SMP, "%s: ipi: %x", __func__, ipi); - ipi_selected(PCPU_GET(other_cpus), ipi); + ipi_selected(other_cpus, ipi); } int ipi_nmi_handler() { - cpumask_t cpumask; + cpuset_t cpumask; /* * As long as there is not a simple way to know about a NMI's @@ -1179,11 +1193,13 @@ ipi_nmi_handler() * the global pending bitword an IPI_STOP_HARD has been issued * and should be handled. */ + sched_pin(); cpumask = PCPU_GET(cpumask); - if ((ipi_nmi_pending & cpumask) == 0) + sched_unpin(); + if (!CPU_OVERLAP(&ipi_nmi_pending, &cpumask)) return (1); - atomic_clear_int(&ipi_nmi_pending, cpumask); + CPU_NAND_ATOMIC(&ipi_nmi_pending, &cpumask); cpustop_handler(); return (0); } @@ -1195,20 +1211,25 @@ ipi_nmi_handler() void cpustop_handler(void) { - int cpu = PCPU_GET(cpuid); - int cpumask = PCPU_GET(cpumask); + cpuset_t cpumask; + int cpu; + + sched_pin(); + cpumask = PCPU_GET(cpumask); + cpu = PCPU_GET(cpuid); + sched_unpin(); savectx(&stoppcbs[cpu]); /* Indicate that we are stopped */ - atomic_set_int(&stopped_cpus, cpumask); + CPU_OR_ATOMIC(&stopped_cpus, &cpumask); /* Wait for restart */ - while (!(started_cpus & cpumask)) + while (!CPU_OVERLAP(&started_cpus, &cpumask)) ia32_pause(); - atomic_clear_int(&started_cpus, cpumask); - atomic_clear_int(&stopped_cpus, cpumask); + CPU_NAND_ATOMIC(&started_cpus, &cpumask); + CPU_NAND_ATOMIC(&stopped_cpus, &cpumask); if (cpu == 0 && cpustop_restartfunc != NULL) { cpustop_restartfunc(); diff --git a/sys/i386/xen/pmap.c b/sys/i386/xen/pmap.c index cf36f5edddb3..74ace613d133 100644 --- a/sys/i386/xen/pmap.c +++ b/sys/i386/xen/pmap.c @@ -422,7 +422,7 @@ pmap_bootstrap(vm_paddr_t firstaddr) #ifdef PAE kernel_pmap->pm_pdpt = (pdpt_entry_t *) (KERNBASE + (u_int)IdlePDPT); #endif - kernel_pmap->pm_active = -1; /* don't allow deactivation */ + CPU_FILL(&kernel_pmap->pm_active); /* don't allow deactivation */ TAILQ_INIT(&kernel_pmap->pm_pvchunk); LIST_INIT(&allpmaps); mtx_init(&allpmaps_lock, "allpmaps", NULL, MTX_SPIN); @@ -802,22 +802,23 @@ pmap_cache_bits(int mode, boolean_t is_pde) void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; CTR2(KTR_PMAP, "pmap_invalidate_page: pmap=%p va=0x%x", pmap, va); sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { invlpg(va); smp_invlpg(va); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) invlpg(va); - if (pmap->pm_active & other_cpus) - smp_masked_invlpg(pmap->pm_active & other_cpus, va); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invlpg(other_cpus, va); } sched_unpin(); PT_UPDATES_FLUSH(); @@ -826,26 +827,26 @@ pmap_invalidate_page(pmap_t pmap, vm_offset_t va) void pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; vm_offset_t addr; CTR3(KTR_PMAP, "pmap_invalidate_page: pmap=%p eva=0x%x sva=0x%x", pmap, sva, eva); sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); smp_invlpg_range(sva, eva); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); - if (pmap->pm_active & other_cpus) - smp_masked_invlpg_range(pmap->pm_active & other_cpus, - sva, eva); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invlpg_range(other_cpus, sva, eva); } sched_unpin(); PT_UPDATES_FLUSH(); @@ -854,21 +855,22 @@ pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) void pmap_invalidate_all(pmap_t pmap) { - cpumask_t cpumask, other_cpus; + cpuset_t cpumask, other_cpus; CTR1(KTR_PMAP, "pmap_invalidate_page: pmap=%p", pmap); sched_pin(); - if (pmap == kernel_pmap || pmap->pm_active == all_cpus) { + if (pmap == kernel_pmap || !CPU_CMP(&pmap->pm_active, &all_cpus)) { invltlb(); smp_invltlb(); } else { cpumask = PCPU_GET(cpumask); other_cpus = PCPU_GET(other_cpus); - if (pmap->pm_active & cpumask) + if (CPU_OVERLAP(&pmap->pm_active, &cpumask)) invltlb(); - if (pmap->pm_active & other_cpus) - smp_masked_invltlb(pmap->pm_active & other_cpus); + CPU_AND(&other_cpus, &pmap->pm_active); + if (!CPU_EMPTY(&other_cpus)) + smp_masked_invltlb(other_cpus); } sched_unpin(); } @@ -893,7 +895,7 @@ pmap_invalidate_page(pmap_t pmap, vm_offset_t va) CTR2(KTR_PMAP, "pmap_invalidate_page: pmap=%p va=0x%x", pmap, va); - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) invlpg(va); PT_UPDATES_FLUSH(); } @@ -907,7 +909,7 @@ pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) CTR3(KTR_PMAP, "pmap_invalidate_range: pmap=%p sva=0x%x eva=0x%x", pmap, sva, eva); - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); PT_UPDATES_FLUSH(); @@ -919,7 +921,7 @@ pmap_invalidate_all(pmap_t pmap) CTR1(KTR_PMAP, "pmap_invalidate_all: pmap=%p", pmap); - if (pmap == kernel_pmap || pmap->pm_active) + if (pmap == kernel_pmap || !CPU_EMPTY(&pmap->pm_active)) invltlb(); } @@ -1449,7 +1451,7 @@ pmap_pinit0(pmap_t pmap) #ifdef PAE pmap->pm_pdpt = (pdpt_entry_t *)(KERNBASE + (vm_offset_t)IdlePDPT); #endif - pmap->pm_active = 0; + CPU_ZERO(&pmap->pm_active); PCPU_SET(curpmap, pmap); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); @@ -1556,7 +1558,7 @@ pmap_pinit(pmap_t pmap) } xen_flush_queue(); vm_page_unlock_queues(); - pmap->pm_active = 0; + CPU_ZERO(&pmap->pm_active); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); @@ -4030,11 +4032,11 @@ pmap_activate(struct thread *td) pmap = vmspace_pmap(td->td_proc->p_vmspace); oldpmap = PCPU_GET(curpmap); #if defined(SMP) - atomic_clear_int(&oldpmap->pm_active, PCPU_GET(cpumask)); - atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask)); + CPU_NAND_ATOMIC(&oldpmap->pm_active, PCPU_PTR(cpumask)); + CPU_OR_ATOMIC(&pmap->pm_active, PCPU_PTR(cpumask)); #else - oldpmap->pm_active &= ~1; - pmap->pm_active |= 1; + CPU_NAND(&oldpmap->pm_active, PCPU_PTR(cpumask)); + CPU_OR(&pmap->pm_active, PCPU_PTR(cpumask)); #endif #ifdef PAE cr3 = vtophys(pmap->pm_pdpt); diff --git a/sys/ia64/ia64/mp_machdep.c b/sys/ia64/ia64/mp_machdep.c index 5804f8cd69b7..015378514687 100644 --- a/sys/ia64/ia64/mp_machdep.c +++ b/sys/ia64/ia64/mp_machdep.c @@ -139,18 +139,18 @@ ia64_ih_rndzvs(struct thread *td, u_int xiv, struct trapframe *tf) static u_int ia64_ih_stop(struct thread *td, u_int xiv, struct trapframe *tf) { - cpumask_t mybit; + cpuset_t mybit; PCPU_INC(md.stats.pcs_nstops); mybit = PCPU_GET(cpumask); savectx(PCPU_PTR(md.pcb)); - atomic_set_int(&stopped_cpus, mybit); - while ((started_cpus & mybit) == 0) + CPU_OR_ATOMIC(&stopped_cpus, &mybit); + while (!CPU_OVERLAP(&started_cpus, &mybit)) cpu_spinwait(); - atomic_clear_int(&started_cpus, mybit); - atomic_clear_int(&stopped_cpus, mybit); + CPU_NAND_ATOMIC(&started_cpus, &mybit); + CPU_NAND_ATOMIC(&stopped_cpus, &mybit); return (0); } @@ -458,12 +458,12 @@ cpu_mp_unleash(void *dummy) * send an IPI to a set of cpus. */ void -ipi_selected(cpumask_t cpus, int ipi) +ipi_selected(cpuset_t cpus, int ipi) { struct pcpu *pc; SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { - if (cpus & pc->pc_cpumask) + if (CPU_OVERLAP(&cpus, &pc->pc_cpumask)) ipi_send(pc, ipi); } } diff --git a/sys/ia64/include/_types.h b/sys/ia64/include/_types.h index 8fc1be2f3873..0c2f5cc38d0b 100644 --- a/sys/ia64/include/_types.h +++ b/sys/ia64/include/_types.h @@ -59,7 +59,6 @@ typedef unsigned long __uint64_t; * Standard type definitions. */ typedef __int32_t __clock_t; /* clock()... */ -typedef unsigned int __cpumask_t; typedef __int64_t __critical_t; typedef double __double_t; typedef float __float_t; diff --git a/sys/ia64/include/smp.h b/sys/ia64/include/smp.h index 26557a712705..d2aff76c3ad5 100644 --- a/sys/ia64/include/smp.h +++ b/sys/ia64/include/smp.h @@ -14,6 +14,8 @@ #ifndef LOCORE +#include <sys/_cpuset.h> + struct pcpu; struct ia64_ap_state { @@ -44,7 +46,7 @@ extern int ia64_ipi_wakeup; void ipi_all_but_self(int ipi); void ipi_cpu(int cpu, u_int ipi); -void ipi_selected(cpumask_t cpus, int ipi); +void ipi_selected(cpuset_t cpus, int ipi); void ipi_send(struct pcpu *, int ipi); #endif /* !LOCORE */ diff --git a/sys/kern/kern_cpuset.c b/sys/kern/kern_cpuset.c index bf9eac79ac8e..9ed19d4d60d9 100644 --- a/sys/kern/kern_cpuset.c +++ b/sys/kern/kern_cpuset.c @@ -617,6 +617,49 @@ out: } /* + * Calculate the ffs() of the cpuset. + */ +int +cpusetobj_ffs(const cpuset_t *set) +{ + size_t i; + int cbit; + + cbit = 0; + for (i = 0; i < _NCPUWORDS; i++) { + if (set->__bits[i] != 0) { + cbit = ffsl(set->__bits[i]); + cbit += i * _NCPUBITS; + break; + } + } + return (cbit); +} + +/* + * Return a string representing a valid layout for a cpuset_t object. + * It expects an incoming buffer at least sized as CPUSETBUFSIZ. + */ +char * +cpusetobj_strprint(char *buf, const cpuset_t *set) +{ + char *tbuf; + size_t i, bytesp, bufsiz; + + tbuf = buf; + bytesp = 0; + bufsiz = CPUSETBUFSIZ; + + for (i = 0; i < (_NCPUWORDS - 1); i++) { + bytesp = snprintf(tbuf, bufsiz, "%lx, ", set->__bits[i]); + bufsiz -= bytesp; + tbuf += bytesp; + } + snprintf(tbuf, bufsiz, "%lx", set->__bits[_NCPUWORDS - 1]); + return (buf); +} + +/* * Apply an anonymous mask to a single thread. */ int @@ -754,11 +797,10 @@ cpuset_init(void *arg) { cpuset_t mask; - CPU_ZERO(&mask); #ifdef SMP - mask.__bits[0] = all_cpus; + mask = all_cpus; #else - mask.__bits[0] = 1; + CPU_SETOF(0, &mask); #endif if (cpuset_modify(cpuset_zero, &mask)) panic("Can't set initial cpuset mask.\n"); diff --git a/sys/kern/kern_pmc.c b/sys/kern/kern_pmc.c index 7532378c7a2d..8d9c7c04a711 100644 --- a/sys/kern/kern_pmc.c +++ b/sys/kern/kern_pmc.c @@ -55,7 +55,7 @@ int (*pmc_hook)(struct thread *td, int function, void *arg) = NULL; int (*pmc_intr)(int cpu, struct trapframe *tf) = NULL; /* Bitmask of CPUs requiring servicing at hardclock time */ -volatile cpumask_t pmc_cpumask; +volatile cpuset_t pmc_cpumask; /* * A global count of SS mode PMCs. When non-zero, this means that @@ -112,7 +112,7 @@ pmc_cpu_is_active(int cpu) { #ifdef SMP return (pmc_cpu_is_present(cpu) && - (hlt_cpus_mask & (1 << cpu)) == 0); + !CPU_ISSET(cpu, &hlt_cpus_mask)); #else return (1); #endif @@ -139,7 +139,7 @@ int pmc_cpu_is_primary(int cpu) { #ifdef SMP - return ((logical_cpus_mask & (1 << cpu)) == 0); + return (!CPU_ISSET(cpu, &logical_cpus_mask)); #else return (1); #endif diff --git a/sys/kern/kern_rmlock.c b/sys/kern/kern_rmlock.c index 7f2b4e7367be..3214e1b55e2f 100644 --- a/sys/kern/kern_rmlock.c +++ b/sys/kern/kern_rmlock.c @@ -263,7 +263,7 @@ _rm_rlock_hard(struct rmlock *rm, struct rm_priotracker *tracker, int trylock) pc = pcpu_find(curcpu); /* Check if we just need to do a proper critical_exit. */ - if (!(pc->pc_cpumask & rm->rm_writecpus)) { + if (!CPU_OVERLAP(&pc->pc_cpumask, &rm->rm_writecpus)) { critical_exit(); return (1); } @@ -325,7 +325,7 @@ _rm_rlock_hard(struct rmlock *rm, struct rm_priotracker *tracker, int trylock) critical_enter(); pc = pcpu_find(curcpu); - rm->rm_writecpus &= ~pc->pc_cpumask; + CPU_NAND(&rm->rm_writecpus, &pc->pc_cpumask); rm_tracker_add(pc, tracker); sched_pin(); critical_exit(); @@ -366,7 +366,8 @@ _rm_rlock(struct rmlock *rm, struct rm_priotracker *tracker, int trylock) * Fast path to combine two common conditions into a single * conditional jump. */ - if (0 == (td->td_owepreempt | (rm->rm_writecpus & pc->pc_cpumask))) + if (0 == (td->td_owepreempt | + CPU_OVERLAP(&rm->rm_writecpus, &pc->pc_cpumask))) return (1); /* We do not have a read token and need to acquire one. */ @@ -429,17 +430,17 @@ _rm_wlock(struct rmlock *rm) { struct rm_priotracker *prio; struct turnstile *ts; - cpumask_t readcpus; + cpuset_t readcpus; if (rm->lock_object.lo_flags & RM_SLEEPABLE) sx_xlock(&rm->rm_lock_sx); else mtx_lock(&rm->rm_lock_mtx); - if (rm->rm_writecpus != all_cpus) { + if (CPU_CMP(&rm->rm_writecpus, &all_cpus)) { /* Get all read tokens back */ - - readcpus = all_cpus & (all_cpus & ~rm->rm_writecpus); + readcpus = all_cpus; + CPU_NAND(&readcpus, &rm->rm_writecpus); rm->rm_writecpus = all_cpus; /* diff --git a/sys/kern/sched_4bsd.c b/sys/kern/sched_4bsd.c index fef9e256f6bb..2fad27ce505c 100644 --- a/sys/kern/sched_4bsd.c +++ b/sys/kern/sched_4bsd.c @@ -156,7 +156,7 @@ static struct runq runq; static struct runq runq_pcpu[MAXCPU]; long runq_length[MAXCPU]; -static cpumask_t idle_cpus_mask; +static cpuset_t idle_cpus_mask; #endif struct pcpuidlestat { @@ -951,7 +951,8 @@ sched_switch(struct thread *td, struct thread *newtd, int flags) if (td->td_flags & TDF_IDLETD) { TD_SET_CAN_RUN(td); #ifdef SMP - idle_cpus_mask &= ~PCPU_GET(cpumask); + /* Spinlock held here, assume no migration. */ + CPU_NAND(&idle_cpus_mask, PCPU_PTR(cpumask)); #endif } else { if (TD_IS_RUNNING(td)) { @@ -1025,7 +1026,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags) #ifdef SMP if (td->td_flags & TDF_IDLETD) - idle_cpus_mask |= PCPU_GET(cpumask); + CPU_OR(&idle_cpus_mask, PCPU_PTR(cpumask)); #endif sched_lock.mtx_lock = (uintptr_t)td; td->td_oncpu = PCPU_GET(cpuid); @@ -1054,7 +1055,8 @@ static int forward_wakeup(int cpunum) { struct pcpu *pc; - cpumask_t dontuse, id, map, map2, me; + cpuset_t dontuse, id, map, map2, me; + int iscpuset; mtx_assert(&sched_lock, MA_OWNED); @@ -1071,32 +1073,38 @@ forward_wakeup(int cpunum) /* * Check the idle mask we received against what we calculated * before in the old version. + * + * Also note that sched_lock is held now, thus no migration is + * expected. */ me = PCPU_GET(cpumask); /* Don't bother if we should be doing it ourself. */ - if ((me & idle_cpus_mask) && (cpunum == NOCPU || me == (1 << cpunum))) + if (CPU_OVERLAP(&me, &idle_cpus_mask) && + (cpunum == NOCPU || CPU_ISSET(cpunum, &me))) return (0); - dontuse = me | stopped_cpus | hlt_cpus_mask; - map2 = 0; + dontuse = me; + CPU_OR(&dontuse, &stopped_cpus); + CPU_OR(&dontuse, &hlt_cpus_mask); + CPU_ZERO(&map2); if (forward_wakeup_use_loop) { SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { id = pc->pc_cpumask; - if ((id & dontuse) == 0 && + if (!CPU_OVERLAP(&id, &dontuse) && pc->pc_curthread == pc->pc_idlethread) { - map2 |= id; + CPU_OR(&map2, &id); } } } if (forward_wakeup_use_mask) { - map = 0; - map = idle_cpus_mask & ~dontuse; + map = idle_cpus_mask; + CPU_NAND(&map, &dontuse); /* If they are both on, compare and use loop if different. */ if (forward_wakeup_use_loop) { - if (map != map2) { + if (CPU_CMP(&map, &map2)) { printf("map != map2, loop method preferred\n"); map = map2; } @@ -1108,18 +1116,22 @@ forward_wakeup(int cpunum) /* If we only allow a specific CPU, then mask off all the others. */ if (cpunum != NOCPU) { KASSERT((cpunum <= mp_maxcpus),("forward_wakeup: bad cpunum.")); - map &= (1 << cpunum); + iscpuset = CPU_ISSET(cpunum, &map); + if (iscpuset == 0) + CPU_ZERO(&map); + else + CPU_SETOF(cpunum, &map); } - if (map) { + if (!CPU_EMPTY(&map)) { forward_wakeups_delivered++; SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { id = pc->pc_cpumask; - if ((map & id) == 0) + if (!CPU_OVERLAP(&map, &id)) continue; if (cpu_idle_wakeup(pc->pc_cpuid)) - map &= ~id; + CPU_NAND(&map, &id); } - if (map) + if (!CPU_EMPTY(&map)) ipi_selected(map, IPI_AST); return (1); } @@ -1135,7 +1147,7 @@ kick_other_cpu(int pri, int cpuid) int cpri; pcpu = pcpu_find(cpuid); - if (idle_cpus_mask & pcpu->pc_cpumask) { + if (CPU_OVERLAP(&idle_cpus_mask, &pcpu->pc_cpumask)) { forward_wakeups_delivered++; if (!cpu_idle_wakeup(cpuid)) ipi_cpu(cpuid, IPI_AST); @@ -1193,6 +1205,7 @@ void sched_add(struct thread *td, int flags) #ifdef SMP { + cpuset_t idle, me, tidlemsk; struct td_sched *ts; int forwarded = 0; int cpu; @@ -1262,11 +1275,20 @@ sched_add(struct thread *td, int flags) kick_other_cpu(td->td_priority, cpu); } else { if (!single_cpu) { - cpumask_t me = PCPU_GET(cpumask); - cpumask_t idle = idle_cpus_mask & me; - if (!idle && ((flags & SRQ_INTR) == 0) && - (idle_cpus_mask & ~(hlt_cpus_mask | me))) + /* + * Thread spinlock is held here, assume no + * migration is possible. + */ + me = PCPU_GET(cpumask); + idle = idle_cpus_mask; + tidlemsk = idle; + CPU_AND(&idle, &me); + CPU_OR(&me, &hlt_cpus_mask); + CPU_NAND(&tidlemsk, &me); + + if (CPU_EMPTY(&idle) && ((flags & SRQ_INTR) == 0) && + !CPU_EMPTY(&tidlemsk)) forwarded = forward_wakeup(cpu); } diff --git a/sys/kern/sched_ule.c b/sys/kern/sched_ule.c index ac18e778ae28..05267f3d6adb 100644 --- a/sys/kern/sched_ule.c +++ b/sys/kern/sched_ule.c @@ -564,7 +564,7 @@ struct cpu_search { #define CPUSET_FOREACH(cpu, mask) \ for ((cpu) = 0; (cpu) <= mp_maxid; (cpu)++) \ - if ((mask) & 1 << (cpu)) + if (CPU_ISSET(cpu, &mask)) static __inline int cpu_search(struct cpu_group *cg, struct cpu_search *low, struct cpu_search *high, const int match); @@ -2650,15 +2650,16 @@ static int sysctl_kern_sched_topology_spec_internal(struct sbuf *sb, struct cpu_group *cg, int indent) { + char cpusetbuf[CPUSETBUFSIZ]; int i, first; sbuf_printf(sb, "%*s<group level=\"%d\" cache-level=\"%d\">\n", indent, "", 1 + indent / 2, cg->cg_level); - sbuf_printf(sb, "%*s <cpu count=\"%d\" mask=\"0x%x\">", indent, "", - cg->cg_count, cg->cg_mask); + sbuf_printf(sb, "%*s <cpu count=\"%d\" mask=\"%s\">", indent, "", + cg->cg_count, cpusetobj_strprint(cpusetbuf, &cg->cg_mask)); first = TRUE; for (i = 0; i < MAXCPU; i++) { - if ((cg->cg_mask & (1 << i)) != 0) { + if (CPU_ISSET(i, &cg->cg_mask)) { if (!first) sbuf_printf(sb, ", "); else diff --git a/sys/kern/subr_kdb.c b/sys/kern/subr_kdb.c index 342c5ca1879d..bb78c00a7591 100644 --- a/sys/kern/subr_kdb.c +++ b/sys/kern/subr_kdb.c @@ -413,7 +413,8 @@ kdb_thr_ctx(struct thread *thr) #if defined(SMP) && defined(KDB_STOPPEDPCB) SLIST_FOREACH(pc, &cpuhead, pc_allcpu) { - if (pc->pc_curthread == thr && (stopped_cpus & pc->pc_cpumask)) + if (pc->pc_curthread == thr && + CPU_OVERLAP(&stopped_cpus, &pc->pc_cpumask)) return (KDB_STOPPEDPCB(pc)); } #endif diff --git a/sys/kern/subr_pcpu.c b/sys/kern/subr_pcpu.c index de5cafccf7f3..9201bb7e8cfc 100644 --- a/sys/kern/subr_pcpu.c +++ b/sys/kern/subr_pcpu.c @@ -87,7 +87,7 @@ pcpu_init(struct pcpu *pcpu, int cpuid, size_t size) KASSERT(cpuid >= 0 && cpuid < MAXCPU, ("pcpu_init: invalid cpuid %d", cpuid)); pcpu->pc_cpuid = cpuid; - pcpu->pc_cpumask = 1 << cpuid; + CPU_SETOF(cpuid, &pcpu->pc_cpumask); cpuid_to_pcpu[cpuid] = pcpu; SLIST_INSERT_HEAD(&cpuhead, pcpu, pc_allcpu); cpu_pcpu_init(pcpu, cpuid, size); diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c index aba6f0e1ea7e..ae061f36c3b5 100644 --- a/sys/kern/subr_smp.c +++ b/sys/kern/subr_smp.c @@ -53,15 +53,15 @@ __FBSDID("$FreeBSD$"); #include "opt_sched.h" #ifdef SMP -volatile cpumask_t stopped_cpus; -volatile cpumask_t started_cpus; -cpumask_t hlt_cpus_mask; -cpumask_t logical_cpus_mask; +volatile cpuset_t stopped_cpus; +volatile cpuset_t started_cpus; +cpuset_t hlt_cpus_mask; +cpuset_t logical_cpus_mask; void (*cpustop_restartfunc)(void); #endif /* This is used in modules that need to work in both SMP and UP. */ -cpumask_t all_cpus; +cpuset_t all_cpus; int mp_ncpus; /* export this for libkvm consumers. */ @@ -199,8 +199,11 @@ forward_signal(struct thread *td) * */ static int -generic_stop_cpus(cpumask_t map, u_int type) +generic_stop_cpus(cpuset_t map, u_int type) { +#ifdef KTR + char cpusetbuf[CPUSETBUFSIZ]; +#endif static volatile u_int stopping_cpu = NOCPU; int i; @@ -215,7 +218,8 @@ generic_stop_cpus(cpumask_t map, u_int type) if (!smp_started) return (0); - CTR2(KTR_SMP, "stop_cpus(%x) with %u type", map, type); + CTR2(KTR_SMP, "stop_cpus(%s) with %u type", + cpusetobj_strprint(cpusetbuf, &map), type); if (stopping_cpu != PCPU_GET(cpuid)) while (atomic_cmpset_int(&stopping_cpu, NOCPU, @@ -227,7 +231,7 @@ generic_stop_cpus(cpumask_t map, u_int type) ipi_selected(map, type); i = 0; - while ((stopped_cpus & map) != map) { + while (!CPU_SUBSET(&stopped_cpus, &map)) { /* spin */ cpu_spinwait(); i++; @@ -244,14 +248,14 @@ generic_stop_cpus(cpumask_t map, u_int type) } int -stop_cpus(cpumask_t map) +stop_cpus(cpuset_t map) { return (generic_stop_cpus(map, IPI_STOP)); } int -stop_cpus_hard(cpumask_t map) +stop_cpus_hard(cpuset_t map) { return (generic_stop_cpus(map, IPI_STOP_HARD)); @@ -259,7 +263,7 @@ stop_cpus_hard(cpumask_t map) #if defined(__amd64__) int -suspend_cpus(cpumask_t map) +suspend_cpus(cpuset_t map) { return (generic_stop_cpus(map, IPI_SUSPEND)); @@ -280,19 +284,22 @@ suspend_cpus(cpumask_t map) * 1: ok */ int -restart_cpus(cpumask_t map) +restart_cpus(cpuset_t map) { +#ifdef KTR + char cpusetbuf[CPUSETBUFSIZ]; +#endif if (!smp_started) return 0; - CTR1(KTR_SMP, "restart_cpus(%x)", map); + CTR1(KTR_SMP, "restart_cpus(%s)", cpusetobj_strprint(cpusetbuf, &map)); /* signal other cpus to restart */ - atomic_store_rel_int(&started_cpus, map); + CPU_COPY_STORE_REL(&map, &started_cpus); /* wait for each to clear its bit */ - while ((stopped_cpus & map) != 0) + while (CPU_OVERLAP(&stopped_cpus, &map)) cpu_spinwait(); return 1; @@ -348,13 +355,13 @@ smp_rendezvous_action(void) } void -smp_rendezvous_cpus(cpumask_t map, +smp_rendezvous_cpus(cpuset_t map, void (* setup_func)(void *), void (* action_func)(void *), void (* teardown_func)(void *), void *arg) { - int i, ncpus = 0; + int curcpumap, i, ncpus = 0; if (!smp_started) { if (setup_func != NULL) @@ -367,11 +374,11 @@ smp_rendezvous_cpus(cpumask_t map, } CPU_FOREACH(i) { - if (((1 << i) & map) != 0) + if (CPU_ISSET(i, &map)) ncpus++; } if (ncpus == 0) - panic("ncpus is 0 with map=0x%x", map); + panic("ncpus is 0 with non-zero map"); /* obtain rendezvous lock */ mtx_lock_spin(&smp_ipi_mtx); @@ -387,10 +394,12 @@ smp_rendezvous_cpus(cpumask_t map, atomic_store_rel_int(&smp_rv_waiters[0], 0); /* signal other processors, which will enter the IPI with interrupts off */ - ipi_selected(map & ~(1 << curcpu), IPI_RENDEZVOUS); + curcpumap = CPU_ISSET(curcpu, &map); + CPU_CLR(curcpu, &map); + ipi_selected(map, IPI_RENDEZVOUS); /* Check if the current CPU is in the map */ - if ((map & (1 << curcpu)) != 0) + if (curcpumap != 0) smp_rendezvous_action(); if (teardown_func == smp_no_rendevous_barrier) @@ -415,6 +424,7 @@ static struct cpu_group group[MAXCPU]; struct cpu_group * smp_topo(void) { + char cpusetbuf[CPUSETBUFSIZ], cpusetbuf2[CPUSETBUFSIZ]; struct cpu_group *top; /* @@ -461,9 +471,10 @@ smp_topo(void) if (top->cg_count != mp_ncpus) panic("Built bad topology at %p. CPU count %d != %d", top, top->cg_count, mp_ncpus); - if (top->cg_mask != all_cpus) - panic("Built bad topology at %p. CPU mask 0x%X != 0x%X", - top, top->cg_mask, all_cpus); + if (CPU_CMP(&top->cg_mask, &all_cpus)) + panic("Built bad topology at %p. CPU mask (%s) != (%s)", + top, cpusetobj_strprint(cpusetbuf, &top->cg_mask), + cpusetobj_strprint(cpusetbuf2, &all_cpus)); return (top); } @@ -488,11 +499,13 @@ static int smp_topo_addleaf(struct cpu_group *parent, struct cpu_group *child, int share, int count, int flags, int start) { - cpumask_t mask; + char cpusetbuf[CPUSETBUFSIZ], cpusetbuf2[CPUSETBUFSIZ]; + cpuset_t mask; int i; - for (mask = 0, i = 0; i < count; i++, start++) - mask |= (1 << start); + CPU_ZERO(&mask); + for (i = 0; i < count; i++, start++) + CPU_SET(start, &mask); child->cg_parent = parent; child->cg_child = NULL; child->cg_children = 0; @@ -502,10 +515,12 @@ smp_topo_addleaf(struct cpu_group *parent, struct cpu_group *child, int share, child->cg_mask = mask; parent->cg_children++; for (; parent != NULL; parent = parent->cg_parent) { - if ((parent->cg_mask & child->cg_mask) != 0) - panic("Duplicate children in %p. mask 0x%X child 0x%X", - parent, parent->cg_mask, child->cg_mask); - parent->cg_mask |= child->cg_mask; + if (CPU_OVERLAP(&parent->cg_mask, &child->cg_mask)) + panic("Duplicate children in %p. mask (%s) child (%s)", + parent, + cpusetobj_strprint(cpusetbuf, &parent->cg_mask), + cpusetobj_strprint(cpusetbuf2, &child->cg_mask)); + CPU_OR(&parent->cg_mask, &child->cg_mask); parent->cg_count += child->cg_count; } @@ -565,20 +580,20 @@ struct cpu_group * smp_topo_find(struct cpu_group *top, int cpu) { struct cpu_group *cg; - cpumask_t mask; + cpuset_t mask; int children; int i; - mask = (1 << cpu); + CPU_SETOF(cpu, &mask); cg = top; for (;;) { - if ((cg->cg_mask & mask) == 0) + if (!CPU_OVERLAP(&cg->cg_mask, &mask)) return (NULL); if (cg->cg_children == 0) return (cg); children = cg->cg_children; for (i = 0, cg = cg->cg_child; i < children; cg++, i++) - if ((cg->cg_mask & mask) != 0) + if (CPU_OVERLAP(&cg->cg_mask, &mask)) break; } return (NULL); @@ -586,7 +601,7 @@ smp_topo_find(struct cpu_group *top, int cpu) #else /* !SMP */ void -smp_rendezvous_cpus(cpumask_t map, +smp_rendezvous_cpus(cpuset_t map, void (*setup_func)(void *), void (*action_func)(void *), void (*teardown_func)(void *), diff --git a/sys/ofed/include/linux/list.h b/sys/ofed/include/linux/list.h index f6f9404307c8..61b42d242a43 100644 --- a/sys/ofed/include/linux/list.h +++ b/sys/ofed/include/linux/list.h @@ -38,6 +38,7 @@ #include <sys/param.h> #include <sys/kernel.h> #include <sys/queue.h> +#include <sys/cpuset.h> #include <sys/lock.h> #include <sys/mutex.h> #include <sys/proc.h> diff --git a/sys/sys/_cpuset.h b/sys/sys/_cpuset.h new file mode 100644 index 000000000000..62a0e9d425d0 --- /dev/null +++ b/sys/sys/_cpuset.h @@ -0,0 +1,52 @@ +/*- + * Copyright (c) 2008, Jeffrey Roberson <jeff@freebsd.org> + * All rights reserved. + * + * Copyright (c) 2008 Nokia Corporation + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice unmodified, this list of conditions, and the following + * disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. + * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, + * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT + * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF + * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#ifndef _SYS__CPUSET_H_ +#define _SYS__CPUSET_H_ + +#ifdef _KERNEL +#define CPU_SETSIZE MAXCPU +#endif + +#define CPU_MAXSIZE (4 * MAXCPU) + +#ifndef CPU_SETSIZE +#define CPU_SETSIZE CPU_MAXSIZE +#endif + +#define _NCPUBITS (sizeof(long) * NBBY) /* bits per mask */ +#define _NCPUWORDS howmany(CPU_SETSIZE, _NCPUBITS) + +typedef struct _cpuset { + long __bits[howmany(CPU_SETSIZE, _NCPUBITS)]; +} cpuset_t; + +#endif /* !_SYS__CPUSET_H_ */ diff --git a/sys/sys/_rmlock.h b/sys/sys/_rmlock.h index 75a159c1082e..15d6c4953e85 100644 --- a/sys/sys/_rmlock.h +++ b/sys/sys/_rmlock.h @@ -45,7 +45,7 @@ LIST_HEAD(rmpriolist,rm_priotracker); struct rmlock { struct lock_object lock_object; - volatile cpumask_t rm_writecpus; + volatile cpuset_t rm_writecpus; LIST_HEAD(,rm_priotracker) rm_activeReaders; union { struct mtx _rm_lock_mtx; diff --git a/sys/sys/cpuset.h b/sys/sys/cpuset.h index 32639913973c..e06df54ab018 100644 --- a/sys/sys/cpuset.h +++ b/sys/sys/cpuset.h @@ -32,22 +32,9 @@ #ifndef _SYS_CPUSET_H_ #define _SYS_CPUSET_H_ -#ifdef _KERNEL -#define CPU_SETSIZE MAXCPU -#endif +#include <sys/_cpuset.h> -#define CPU_MAXSIZE (4 * MAXCPU) - -#ifndef CPU_SETSIZE -#define CPU_SETSIZE CPU_MAXSIZE -#endif - -#define _NCPUBITS (sizeof(long) * NBBY) /* bits per mask */ -#define _NCPUWORDS howmany(CPU_SETSIZE, _NCPUBITS) - -typedef struct _cpuset { - long __bits[howmany(CPU_SETSIZE, _NCPUBITS)]; -} cpuset_t; +#define CPUSETBUFSIZ ((2 + sizeof(long) * 2) * _NCPUWORDS) #define __cpuset_mask(n) ((long)1 << ((n) % _NCPUBITS)) #define CPU_CLR(n, p) ((p)->__bits[(n)/_NCPUBITS] &= ~__cpuset_mask(n)) @@ -66,6 +53,11 @@ typedef struct _cpuset { (p)->__bits[__i] = -1; \ } while (0) +#define CPU_SETOF(n, p) do { \ + CPU_ZERO(p); \ + ((p)->__bits[(n)/_NCPUBITS] = __cpuset_mask(n)); \ +} while (0) + /* Is p empty. */ #define CPU_EMPTY(p) __extension__ ({ \ __size_t __i; \ @@ -75,6 +67,15 @@ typedef struct _cpuset { __i == _NCPUWORDS; \ }) +/* Is p full set. */ +#define CPU_ISFULLSET(p) __extension__ ({ \ + __size_t __i; \ + for (__i = 0; __i < _NCPUWORDS; __i++) \ + if ((p)->__bits[__i] != (long)-1) \ + break; \ + __i == _NCPUWORDS; \ +}) + /* Is c a subset of p. */ #define CPU_SUBSET(p, c) __extension__ ({ \ __size_t __i; \ @@ -124,6 +125,35 @@ typedef struct _cpuset { (d)->__bits[__i] &= ~(s)->__bits[__i]; \ } while (0) +#ifdef _KERNEL +#define CPU_CLR_ATOMIC(n, p) \ + atomic_clear_long(&(p)->__bits[(n)/_NCPUBITS], __cpuset_mask(n)) + +#define CPU_SET_ATOMIC(n, p) \ + atomic_set_long(&(p)->__bits[(n)/_NCPUBITS], __cpuset_mask(n)) + +#define CPU_OR_ATOMIC(d, s) do { \ + __size_t __i; \ + for (__i = 0; __i < _NCPUWORDS; __i++) \ + atomic_set_long(&(d)->__bits[__i], \ + (s)->__bits[__i]); \ +} while (0) + +#define CPU_NAND_ATOMIC(d, s) do { \ + __size_t __i; \ + for (__i = 0; __i < _NCPUWORDS; __i++) \ + atomic_clear_long(&(d)->__bits[__i], \ + (s)->__bits[__i]); \ +} while (0) + +#define CPU_COPY_STORE_REL(f, t) do { \ + __size_t __i; \ + for (__i = 0; __i < _NCPUWORDS; __i++) \ + atomic_store_rel_long(&(t)->__bits[__i], \ + (f)->__bits[__i]); \ +} while (0) +#endif /* !_KERNEL */ + /* * Valid cpulevel_t values. */ @@ -184,6 +214,8 @@ void cpuset_rel(struct cpuset *); int cpuset_setthread(lwpid_t id, cpuset_t *); int cpuset_create_root(struct prison *, struct cpuset **); int cpuset_setproc_update_set(struct proc *, struct cpuset *); +int cpusetobj_ffs(const cpuset_t *); +char *cpusetobj_strprint(char *, const cpuset_t *); #else __BEGIN_DECLS diff --git a/sys/sys/pcpu.h b/sys/sys/pcpu.h index ad1cf33e05b2..1b7f24f39117 100644 --- a/sys/sys/pcpu.h +++ b/sys/sys/pcpu.h @@ -37,6 +37,7 @@ #error "no assembler-serviceable parts inside" #endif +#include <sys/_cpuset.h> #include <sys/queue.h> #include <sys/vmmeter.h> #include <sys/resource.h> @@ -162,8 +163,8 @@ struct pcpu { uint64_t pc_switchtime; /* cpu_ticks() at last csw */ int pc_switchticks; /* `ticks' at last csw */ u_int pc_cpuid; /* This cpu number */ - cpumask_t pc_cpumask; /* This cpu mask */ - cpumask_t pc_other_cpus; /* Mask of all other cpus */ + cpuset_t pc_cpumask; /* This cpu mask */ + cpuset_t pc_other_cpus; /* Mask of all other cpus */ SLIST_ENTRY(pcpu) pc_allcpu; struct lock_list_entry *pc_spinlocks; #ifdef KTR diff --git a/sys/sys/pmckern.h b/sys/sys/pmckern.h index 3e8c1ef3fb68..796c4cadab77 100644 --- a/sys/sys/pmckern.h +++ b/sys/sys/pmckern.h @@ -76,7 +76,7 @@ extern int (*pmc_intr)(int _cpu, struct trapframe *_frame); extern struct sx pmc_sx; /* Per-cpu flags indicating availability of sampling data */ -extern volatile cpumask_t pmc_cpumask; +extern volatile cpuset_t pmc_cpumask; /* Count of system-wide sampling PMCs in existence */ extern volatile int pmc_ss_count; @@ -122,7 +122,7 @@ do { \ #define PMC_SYSTEM_SAMPLING_ACTIVE() (pmc_ss_count > 0) /* Check if a CPU has recorded samples. */ -#define PMC_CPU_HAS_SAMPLES(C) (__predict_false(pmc_cpumask & (1 << (C)))) +#define PMC_CPU_HAS_SAMPLES(C) (__predict_false(CPU_ISSET(C, &pmc_cpumask))) /* * Helper functions. diff --git a/sys/sys/smp.h b/sys/sys/smp.h index 544cb9575f91..410adfa827e3 100644 --- a/sys/sys/smp.h +++ b/sys/sys/smp.h @@ -16,6 +16,8 @@ #ifndef LOCORE +#include <sys/cpuset.h> + /* * Topology of a NUMA or HTT system. * @@ -32,7 +34,7 @@ struct cpu_group { struct cpu_group *cg_parent; /* Our parent group. */ struct cpu_group *cg_child; /* Optional children groups. */ - cpumask_t cg_mask; /* Mask of cpus in this group. */ + cpuset_t cg_mask; /* Mask of cpus in this group. */ int8_t cg_count; /* Count of cpus in this group. */ int8_t cg_children; /* Number of children groups. */ int8_t cg_level; /* Shared cache level. */ @@ -71,10 +73,10 @@ struct cpu_group *smp_topo_find(struct cpu_group *top, int cpu); extern void (*cpustop_restartfunc)(void); extern int smp_active; extern int smp_cpus; -extern volatile cpumask_t started_cpus; -extern volatile cpumask_t stopped_cpus; -extern cpumask_t hlt_cpus_mask; -extern cpumask_t logical_cpus_mask; +extern volatile cpuset_t started_cpus; +extern volatile cpuset_t stopped_cpus; +extern cpuset_t hlt_cpus_mask; +extern cpuset_t logical_cpus_mask; #endif /* SMP */ extern u_int mp_maxid; @@ -82,14 +84,14 @@ extern int mp_maxcpus; extern int mp_ncpus; extern volatile int smp_started; -extern cpumask_t all_cpus; +extern cpuset_t all_cpus; /* * Macro allowing us to determine whether a CPU is absent at any given * time, thus permitting us to configure sparse maps of cpuid-dependent * (per-CPU) structures. */ -#define CPU_ABSENT(x_cpu) ((all_cpus & (1 << (x_cpu))) == 0) +#define CPU_ABSENT(x_cpu) (!CPU_ISSET(x_cpu, &all_cpus)) /* * Macros to iterate over non-absent CPUs. CPU_FOREACH() takes an @@ -158,11 +160,11 @@ void cpu_mp_setmaxid(void); void cpu_mp_start(void); void forward_signal(struct thread *); -int restart_cpus(cpumask_t); -int stop_cpus(cpumask_t); -int stop_cpus_hard(cpumask_t); +int restart_cpus(cpuset_t); +int stop_cpus(cpuset_t); +int stop_cpus_hard(cpuset_t); #if defined(__amd64__) -int suspend_cpus(cpumask_t); +int suspend_cpus(cpuset_t); #endif void smp_rendezvous_action(void); extern struct mtx smp_ipi_mtx; @@ -173,7 +175,7 @@ void smp_rendezvous(void (*)(void *), void (*)(void *), void (*)(void *), void *arg); -void smp_rendezvous_cpus(cpumask_t, +void smp_rendezvous_cpus(cpuset_t, void (*)(void *), void (*)(void *), void (*)(void *), diff --git a/sys/sys/types.h b/sys/sys/types.h index c747d166e81d..5cc005d22e19 100644 --- a/sys/sys/types.h +++ b/sys/sys/types.h @@ -143,7 +143,6 @@ typedef __clockid_t clockid_t; #define _CLOCKID_T_DECLARED #endif -typedef __cpumask_t cpumask_t; typedef __critical_t critical_t; /* Critical section value */ typedef __int64_t daddr_t; /* disk address */ diff --git a/sys/x86/x86/local_apic.c b/sys/x86/x86/local_apic.c index 804c411b951c..e81d2fe8af4d 100644 --- a/sys/x86/x86/local_apic.c +++ b/sys/x86/x86/local_apic.c @@ -807,7 +807,7 @@ lapic_handle_timer(struct trapframe *frame) * and unlike other schedulers it actually schedules threads to * those CPUs. */ - if ((hlt_cpus_mask & (1 << PCPU_GET(cpuid))) != 0) + if (CPU_ISSET(PCPU_GET(cpuid), &hlt_cpus_mask)) return; #endif |