From: Nicholas Piggin <npiggin@xxxxxxxxx> Subject: mm/large system hash: clear hashdist when only one node with memory is booted CONFIG_NUMA on 64-bit CPUs currently enables hashdist unconditionally even when booting on single node machines. This causes the large system hashes to be allocated with vmalloc, and mapped with small pages. This change clears hashdist if only one node has come up with memory. This results in the important large inode and dentry hashes using memblock allocations. All others are within 4MB size up to about 128GB of RAM, which allows them to be allocated from the linear map on most non-NUMA images. Other big hashes like futex and TCP should eventually be moved over to the same style of allocation as those vfs caches that use HASH_EARLY if !hashdist, so they don't exceed MAX_ORDER on very large non-NUMA images. This brings dTLB misses for linux kernel tree `git diff` from ~45,000 to ~8,000 on a Kaby Lake KVM guest with 8MB dentry hash and mitigations=off (performance is in the noise, under 1% difference, page tables are likely to be well cached for this workload). Link: http://lkml.kernel.org/r/20190605144814.29319-2-npiggin@xxxxxxxxx Signed-off-by: Nicholas Piggin <npiggin@xxxxxxxxx> Reviewed-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- mm/page_alloc.c | 31 ++++++++++++++++++------------- 1 file changed, 18 insertions(+), 13 deletions(-) --- a/mm/page_alloc.c~mm-large-system-hash-clear-hashdist-when-only-one-node-with-memory-is-booted +++ a/mm/page_alloc.c @@ -7534,10 +7534,28 @@ static int page_alloc_cpu_dead(unsigned return 0; } +#ifdef CONFIG_NUMA +int hashdist = HASHDIST_DEFAULT; + +static int __init set_hashdist(char *str) +{ + if (!str) + return 0; + hashdist = simple_strtoul(str, &str, 0); + return 1; +} +__setup("hashdist=", set_hashdist); +#endif + void __init page_alloc_init(void) { int ret; +#ifdef CONFIG_NUMA + if (num_node_state(N_MEMORY) == 1) + hashdist = 0; +#endif + ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD, "mm/page_alloc:dead", NULL, page_alloc_cpu_dead); @@ -7922,19 +7940,6 @@ out: return ret; } -#ifdef CONFIG_NUMA -int hashdist = HASHDIST_DEFAULT; - -static int __init set_hashdist(char *str) -{ - if (!str) - return 0; - hashdist = simple_strtoul(str, &str, 0); - return 1; -} -__setup("hashdist=", set_hashdist); -#endif - #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES /* * Returns the number of pages that arch has reserved but _
