The patch titled genapic: always use physical delivery mode on > 8 CPUs has been added to the -mm tree. Its filename is genapic-always-use-physical-delivery-mode-on-8-cpus.patch See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find out what to do about this ------------------------------------------------------ Subject: genapic: always use physical delivery mode on > 8 CPUs From: Ingo Molnar <mingo@xxxxxxx> Remove clustered APIC mode. There's little point in the use of clustered APIC mode, broadcasting is limited to within the cluster only, and chipsets have bugs in this area as well. So default to physical APIC mode when the CPU count is large, and default to logical APIC mode when the CPU count is 8 or smaller. (this patch only removes the use of genapic_cluster and cleans up the resulting genapic.c file - removal of all remaining traces of clustered mode will be done by another patch.) Signed-off-by: Ingo Molnar <mingo@xxxxxxx> Cc: Suresh Siddha <suresh.b.siddha@xxxxxxxxx> Cc: Andi Kleen <ak@xxxxxxx> Cc: "Li, Shaohua" <shaohua.li@xxxxxxxxx> Cc: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxx> --- arch/x86_64/kernel/genapic.c | 69 +++++++-------------------------- include/asm-x86_64/genapic.h | 4 + 2 files changed, 19 insertions(+), 54 deletions(-) diff -puN arch/x86_64/kernel/genapic.c~genapic-always-use-physical-delivery-mode-on-8-cpus arch/x86_64/kernel/genapic.c --- a/arch/x86_64/kernel/genapic.c~genapic-always-use-physical-delivery-mode-on-8-cpus +++ a/arch/x86_64/kernel/genapic.c @@ -11,26 +11,24 @@ #include <linux/threads.h> #include <linux/cpumask.h> #include <linux/string.h> +#include <linux/module.h> #include <linux/kernel.h> #include <linux/ctype.h> #include <linux/init.h> -#include <linux/module.h> #include <asm/smp.h> #include <asm/ipi.h> -#if defined(CONFIG_ACPI) +#ifdef CONFIG_ACPI #include <acpi/acpi_bus.h> #endif /* which logical CPU number maps to which CPU (physical APIC ID) */ -u8 x86_cpu_to_apicid[NR_CPUS] __read_mostly = { [0 ... NR_CPUS-1] = BAD_APICID }; +u8 x86_cpu_to_apicid[NR_CPUS] __read_mostly + = { [0 ... NR_CPUS-1] = BAD_APICID }; EXPORT_SYMBOL(x86_cpu_to_apicid); -u8 x86_cpu_to_log_apicid[NR_CPUS] = { [0 ... NR_CPUS-1] = BAD_APICID }; -extern struct genapic apic_cluster; -extern struct genapic apic_flat; -extern struct genapic apic_physflat; +u8 x86_cpu_to_log_apicid[NR_CPUS] = { [0 ... NR_CPUS-1] = BAD_APICID }; struct genapic __read_mostly *genapic = &apic_flat; @@ -39,73 +37,38 @@ struct genapic __read_mostly *genapic = */ void __init clustered_apic_check(void) { - u8 id, clusters, max_cluster, cluster_cnt[NUM_APIC_CLUSTERS]; - int i, max_apic = 0; + unsigned int i, max_apic = 0; + u8 id; #ifdef CONFIG_ACPI /* - * Some x86_64 machines use physical APIC mode regardless of how many - * procs/clusters are present (x86_64 ES7000 is an example). + * Quirk: some x86_64 machines can only use physical APIC mode + * regardless of how many processors are present (x86_64 ES7000 + * is an example). */ if (acpi_fadt.revision > FADT2_REVISION_ID && - acpi_fadt.force_apic_physical_destination_mode) { - genapic = &apic_cluster; - goto print; - } + acpi_fadt.force_apic_physical_destination_mode) + goto force_physflat; #endif - memset(cluster_cnt, 0, sizeof(cluster_cnt)); for (i = 0; i < NR_CPUS; i++) { id = bios_cpu_apicid[i]; if (id == BAD_APICID) continue; if (id > max_apic) max_apic = id; - cluster_cnt[APIC_CLUSTERID(id)]++; - } - - /* - * Don't use clustered mode on AMD platforms, default - * to flat logical mode. - */ - if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) { - /* - * Switch to physical flat mode if more than 8 APICs - * (In the case of 8 CPUs APIC ID goes from 0 to 7): - */ - if (max_apic >= 8) - genapic = &apic_physflat; - goto print; - } - - clusters = 0; - max_cluster = 0; - - for (i = 0; i < NUM_APIC_CLUSTERS; i++) { - if (cluster_cnt[i] > 0) { - ++clusters; - if (cluster_cnt[i] > max_cluster) - max_cluster = cluster_cnt[i]; - } } - /* - * If we have clusters <= 1 and CPUs <= 8 in cluster 0, then flat mode, - * else if max_cluster <= 4 and cluster_cnt[15] == 0, clustered logical - * else physical mode. - * (We don't use lowest priority delivery + HW APIC IRQ steering, so - * can ignore the clustered logical case and go straight to physical.) - */ - if (clusters <= 1 && max_cluster <= 8 && cluster_cnt[0] == max_cluster) + if (max_apic < 8) genapic = &apic_flat; else - genapic = &apic_cluster; +force_physflat: + genapic = &apic_physflat; -print: printk(KERN_INFO "Setting APIC routing to %s\n", genapic->name); } -/* Same for both flat and clustered. */ +/* Same for both flat and physical. */ void send_IPI_self(int vector) { diff -puN include/asm-x86_64/genapic.h~genapic-always-use-physical-delivery-mode-on-8-cpus include/asm-x86_64/genapic.h --- a/include/asm-x86_64/genapic.h~genapic-always-use-physical-delivery-mode-on-8-cpus +++ a/include/asm-x86_64/genapic.h @@ -29,7 +29,9 @@ struct genapic { unsigned int (*phys_pkg_id)(int index_msb); }; - extern struct genapic *genapic; +extern struct genapic apic_flat; +extern struct genapic apic_physflat; + #endif _ Patches currently in -mm which might be from mingo@xxxxxxx are add-bottom_half.h.patch git-scsi-misc.patch revert-x86_64-mm-add-genapic_force.patch fix-x86_64-mm-fix-the-irqbalance-quirk-for-e7320-e7520-e7525.patch cleanup-arch-i386-kernel-smpbootcsmp_tune_scheduling.patch convert-i386-pda-code-to-use-%fs.patch x86_64-check-vector-in-setup_ioapic_dest-to-verify-if-need-setup_io_apic_irq.patch i386-kernel-smpc-dont-use-set_irq_regs.patch genapic-optimize-fix-apic-mode-setup-2.patch genapic-always-use-physical-delivery-mode-on-8-cpus.patch genapic-remove-es7000-workaround.patch genapic-remove-clustered-apic-mode.patch genapic-default-to-physical-mode-on-hotplug-cpu-kernels.patch convert-pm_sem-to-a-mutex.patch convert-pm_sem-to-a-mutex-fix.patch lockdep-annotate-nfs-nfsd-in-kernel-sockets.patch lockdep-annotate-nfs-nfsd-in-kernel-sockets-tidy.patch sysrq-x-show-blocked-tasks.patch lockdep-internal-locking-fixes.patch lockdep-misc-fixes-in-lockdepc.patch cpuset-remove-sched-domain-hooks-from-cpusets.patch binfmt_elf-randomize-pie-binaries.patch disable-clone_child_cleartid-for-abnormal-exit.patch lockdep-annotate-bcsp-driver.patch lockdep-print-current-locks-on-in_atomic-warnings.patch lockdep-name-some-old-style-locks.patch sleep-profiling.patch sleep-profiling-fixes.patch sleep-profiling-fix.patch debug-workqueue-locking-sanity.patch debug-workqueue-locking-sanity-fix.patch retries-in-ext3_prepare_write-violate-ordering-requirements.patch ktime-fix-signed--unsigned-mismatch-in-ktime_to_ns.patch kprobes-enable-booster-on-the-preemptible-kernel.patch declare-smp_call_function_single-in-generic-code.patch up-smp_call_function_single-should-disable-interrupts.patch up-smp_call_function_single-should-disable-interrupts-fix.patch smp_call_function_single-check-that-local-interrupts-are-enabled.patch hotplug-cpu-clean-up-hotcpu_notifier-use.patch remove-kernel-lockdepclockdep_internal.patch lockdep-more-chains.patch lockdep-show-more-details-about-self-test-failures.patch softirq-remove-bug_ons-which-can-incorrectly-trigger.patch futex-remove-unneeded-barrier.patch remove-the-old-bd_mutex-lockdep-annotation.patch new-bd_mutex-lockdep-annotation.patch remove-lock_key-approach-to-managing-nested-bd_mutex-locks.patch simplify-some-aspects-of-bd_mutex-nesting.patch use-mutex_lock_nested-for-bd_mutex-to-avoid-lockdep-warning.patch avoid-lockdep-warning-in-md.patch fix-generic-warn_on-message.patch schedc-correct-comment-for-this_rq_lock-routine.patch sched-fix-migration-cost-estimator.patch sched-domain-move-sched-group-allocations-to-percpu-area.patch move_task_off_dead_cpu-should-be-called-with-disabled-ints.patch sched-domain-increase-the-smt-busy-rebalance-interval.patch sched-avoid-taking-rq-lock-in-wake_priority_sleeper.patch sched-remove-staggering-of-load-balancing.patch sched-disable-interrupts-for-locking-in-load_balance.patch sched-extract-load-calculation-from-rebalance_tick.patch sched-move-idle-status-calculation-into-rebalance_tick.patch sched-use-softirq-for-load-balancing.patch sched-call-tasklet-less-frequently.patch sched-add-option-to-serialize-load-balancing.patch sched-add-option-to-serialize-load-balancing-fix.patch sched-improve-migration-accuracy.patch sched-improve-migration-accuracy-tidy.patch sched-decrease-number-of-load-balances.patch mm-only-sched-add-a-few-scheduler-event-counters.patch sched-optimize-activate_task-for-rt-task.patch kernel-schedc-whitespace-cleanups.patch kernel-schedc-whitespace-cleanups-more.patch sched-add-above-background-load-function.patch mm-implement-swap-prefetching.patch mm-implement-swap-prefetching-use-ctl_unnumbered.patch sched-cleanup-remove-task_t-convert-to-struct-task_struct-prefetch.patch gtod-exponential-update_wall_time.patch gtod-persistent-clock-support-core.patch gtod-persistent-clock-support-i386.patch time-uninline-jiffiesh.patch time-uninline-jiffiesh-fix.patch time-fix-msecs_to_jiffies-bug.patch time-fix-timeout-overflow.patch cleanup-uninline-irq_enter-and-move-it-into-a-function.patch dynticks-extend-next_timer_interrupt-to-use-a-reference-jiffie.patch dynticks-extend-next_timer_interrupt-to-use-a-reference-jiffie-remove-incorrect-warning-in-kernel-timerc.patch hrtimers-namespace-and-enum-cleanup.patch hrtimers-clean-up-locking.patch hrtimers-clean-up-locking-fix.patch updated-hrtimers-state-tracking.patch updated-hrtimers-clean-up-callback-tracking.patch updated-hrtimers-move-and-add-documentation.patch updated-add-a-framework-to-manage-clock-event-devices.patch updated-acpi-include-apich.patch updated-acpi-keep-track-of-timer-broadcast.patch updated-acpi-add-state-propagation-for-dynamic-broadcasting.patch updated-i386-cleanup-apic-code.patch updated-i386-convert-to-clock-event-devices.patch updated-pm_timer-allow-early-access-and-move-externs-to-a-header-file.patch updated-i386-rework-local-apic-calibration.patch updated-high-res-timers-core.patch updated-high-res-timers-core-high-res-timers-do-itimer-rearming-in-process-context.patch updated-gtod-mark-tsc-unusable-for-highres-timers.patch high-res-timers-utilize-tsc-clocksource-again.patch high-res-timers-utilize-tsc-clocksource-again-fix.patch updated-dynticks-core-code.patch updated-dynticks-core-code-fix-resume-bug.patch updated-dyntick-add-nohz-stats-to-proc-stat.patch updated-dynticks-i386-arch-code.patch updated-dynticks-fix-nmi-watchdog.patch updated-high-res-timers-dynticks-enable-i386-support.patch updated-debugging-feature-timer-stats.patch clockevents-core-check-for-clock-event-device-handler-being-non-null-before-calling-it.patch round_jiffies-infrastructure.patch round_jiffies-infrastructure-fix.patch clocksource-add-usage-of-config_sysfs.patch clocksource-small-cleanup-2.patch clocksource-small-cleanup-2-fix.patch clocksource-small-acpi_pm-cleanup.patch kvm-amd-svm-implementation-more-i386-fixes.patch detect-atomic-counter-underflows.patch debug-shared-irqs.patch make-frame_pointer-default=y.patch mutex-subsystem-synchro-test-module.patch vdso-print-fatal-signals.patch vdso-improve-print_fatal_signals-support-by-adding-memory-maps.patch vdso-print-fatal-signals-use-ctl_unnumbered.patch lockdep-show-held-locks-when-showing-a-stackdump.patch lockdep-show-held-locks-when-showing-a-stackdump-fix.patch lockdep-show-held-locks-when-showing-a-stackdump-fix-2.patch - To unsubscribe from this list: send the line "unsubscribe mm-commits" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html