From 0437e109e1841607f2988891eaa36c531c6aa6ac Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 9 Jul 2007 18:51:57 +0200 Subject: sched: zap the migration init / cache-hot balancing code the SMP load-balancer uses the boot-time migration-cost estimation code to attempt to improve the quality of balancing. The reason for this code is that the discrete priority queues do not preserve the order of scheduling accurately, so the load-balancer skips tasks that were running on a CPU 'recently'. this code is fundamental fragile: the boot-time migration cost detector doesnt really work on systems that had large L3 caches, it caused boot delays on large systems and the whole cache-hot concept made the balancing code pretty undeterministic as well. (and hey, i wrote most of it, so i can say it out loud that it sucks ;-) under CFS the same purpose of cache affinity can be achieved without any special cache-hot special-case: tasks are sorted in the 'timeline' tree and the SMP balancer picks tasks from the left side of the tree, thus the most cache-cold task is balanced automatically. Signed-off-by: Ingo Molnar --- arch/i386/kernel/smpboot.c | 12 ------------ arch/ia64/kernel/setup.c | 6 ------ arch/mips/kernel/smp.c | 11 ----------- arch/sparc/kernel/smp.c | 10 ---------- arch/sparc64/kernel/smp.c | 27 --------------------------- 5 files changed, 66 deletions(-) (limited to 'arch') diff --git a/arch/i386/kernel/smpboot.c b/arch/i386/kernel/smpboot.c index 88baed1e7e..0b2954534b 100644 --- a/arch/i386/kernel/smpboot.c +++ b/arch/i386/kernel/smpboot.c @@ -941,17 +941,6 @@ exit: } #endif -static void smp_tune_scheduling(void) -{ - if (cpu_khz) { - /* cache size in kB */ - long cachesize = boot_cpu_data.x86_cache_size; - - if (cachesize > 0) - max_cache_size = cachesize * 1024; - } -} - /* * Cycle through the processors sending APIC IPIs to boot each. */ @@ -980,7 +969,6 @@ static void __init smp_boot_cpus(unsigned int max_cpus) x86_cpu_to_apicid[0] = boot_cpu_physical_apicid; current_thread_info()->cpu = 0; - smp_tune_scheduling(); set_cpu_sibling_map(0); diff --git a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c index eaa6a24bc0..188fb73c68 100644 --- a/arch/ia64/kernel/setup.c +++ b/arch/ia64/kernel/setup.c @@ -805,7 +805,6 @@ static void __cpuinit get_max_cacheline_size (void) { unsigned long line_size, max = 1; - unsigned int cache_size = 0; u64 l, levels, unique_caches; pal_cache_config_info_t cci; s64 status; @@ -835,8 +834,6 @@ get_max_cacheline_size (void) line_size = 1 << cci.pcci_line_size; if (line_size > max) max = line_size; - if (cache_size < cci.pcci_cache_size) - cache_size = cci.pcci_cache_size; if (!cci.pcci_unified) { status = ia64_pal_cache_config_info(l, /* cache_type (instruction)= */ 1, @@ -853,9 +850,6 @@ get_max_cacheline_size (void) ia64_i_cache_stride_shift = cci.pcci_stride; } out: -#ifdef CONFIG_SMP - max_cache_size = max(max_cache_size, cache_size); -#endif if (max > ia64_max_cacheline_size) ia64_max_cacheline_size = max; } diff --git a/arch/mips/kernel/smp.c b/arch/mips/kernel/smp.c index 67edfa7ed9..a1b017f2db 100644 --- a/arch/mips/kernel/smp.c +++ b/arch/mips/kernel/smp.c @@ -51,16 +51,6 @@ int __cpu_logical_map[NR_CPUS]; /* Map logical to physical */ EXPORT_SYMBOL(phys_cpu_present_map); EXPORT_SYMBOL(cpu_online_map); -/* This happens early in bootup, can't really do it better */ -static void smp_tune_scheduling (void) -{ - struct cache_desc *cd = ¤t_cpu_data.scache; - unsigned long cachesize = cd->linesz * cd->sets * cd->ways; - - if (cachesize > max_cache_size) - max_cache_size = cachesize; -} - extern void __init calibrate_delay(void); extern ATTRIB_NORET void cpu_idle(void); @@ -228,7 +218,6 @@ void __init smp_prepare_cpus(unsigned int max_cpus) { init_new_context(current, &init_mm); current_thread_info()->cpu = 0; - smp_tune_scheduling(); plat_prepare_cpus(max_cpus); #ifndef CONFIG_HOTPLUG_CPU cpu_present_map = cpu_possible_map; diff --git a/arch/sparc/kernel/smp.c b/arch/sparc/kernel/smp.c index 4d9ad59031..4fea3ac7bf 100644 --- a/arch/sparc/kernel/smp.c +++ b/arch/sparc/kernel/smp.c @@ -68,16 +68,6 @@ void __cpuinit smp_store_cpu_info(int id) cpu_data(id).prom_node = cpu_node; cpu_data(id).mid = cpu_get_hwmid(cpu_node); - /* this is required to tune the scheduler correctly */ - /* is it possible to have CPUs with different cache sizes? */ - if (id == boot_cpu_id) { - int cache_line,cache_nlines; - cache_line = 0x20; - cache_line = prom_getintdefault(cpu_node, "ecache-line-size", cache_line); - cache_nlines = 0x8000; - cache_nlines = prom_getintdefault(cpu_node, "ecache-nlines", cache_nlines); - max_cache_size = cache_line * cache_nlines; - } if (cpu_data(id).mid < 0) panic("No MID found for CPU%d at node 0x%08d", id, cpu_node); } diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c index 4dcd7d0b60..40e40f968d 100644 --- a/arch/sparc64/kernel/smp.c +++ b/arch/sparc64/kernel/smp.c @@ -1163,32 +1163,6 @@ int setup_profiling_timer(unsigned int multiplier) return -EINVAL; } -static void __init smp_tune_scheduling(void) -{ - unsigned int smallest = ~0U; - int i; - - for (i = 0; i < NR_CPUS; i++) { - unsigned int val = cpu_data(i).ecache_size; - - if (val && val < smallest) - smallest = val; - } - - /* Any value less than 256K is nonsense. */ - if (smallest < (256U * 1024U)) - smallest = 256 * 1024; - - max_cache_size = smallest; - - if (smallest < 1U * 1024U * 1024U) - printk(KERN_INFO "Using max_cache_size of %uKB\n", - smallest / 1024U); - else - printk(KERN_INFO "Using max_cache_size of %uMB\n", - smallest / 1024U / 1024U); -} - /* Constrain the number of cpus to max_cpus. */ void __init smp_prepare_cpus(unsigned int max_cpus) { @@ -1206,7 +1180,6 @@ void __init smp_prepare_cpus(unsigned int max_cpus) } cpu_data(boot_cpu_id).udelay_val = loops_per_jiffy; - smp_tune_scheduling(); } void __devinit smp_prepare_boot_cpu(void) -- cgit v1.2.2