From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758878AbXGMNWY (ORCPT ); Fri, 13 Jul 2007 09:22:24 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1758890AbXGMNVf (ORCPT ); Fri, 13 Jul 2007 09:21:35 -0400 Received: from nz-out-0506.google.com ([64.233.162.239]:51966 "EHLO nz-out-0506.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758181AbXGMNVc (ORCPT ); Fri, 13 Jul 2007 09:21:32 -0400 DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:sender:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references:x-google-sender-auth; b=Bd0vqWs2iJRWfYMjl+lptNFhpLXNHIeZ3edjLMT77Poo0MKbqt66k4ZT7KCXaIoTzsng7Qb8iKYUKHlo4CgHufQTDyD722qpzY8PCAbf6euArwuPMhU7fioqTatSRNkOj7j+VqXXfGoDdZYtR1f5zEUH3wpyvwX+qQIAtxJ8nSc= Message-ID: <84144f020707130621g1d3618e0r85545f3aa26c7466@mail.gmail.com> Date: Fri, 13 Jul 2007 16:21:30 +0300 From: "Pekka Enberg" To: "Kirill Korotaev" Subject: Re: [PATCH] replace kmem_cache_alloc with kmem_cache_zalloc to remove some following zero initializations. Cc: "rae l" , trivial@kernel.org, Denis , linux-kernel@vger.kernel.org In-Reply-To: <46974193.2020302@sw.ru> MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Content-Disposition: inline References: <1184290905.5475.5.camel@tux.homenetwork> <4697357A.10605@sw.ru> <91b13c310707130138p6cffd0b4va34eec320d66ab44@mail.gmail.com> <46974193.2020302@sw.ru> X-Google-Sender-Auth: e66ddfd5373c32d7 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Hi Kirill, On 7/13/07, Kirill Korotaev wrote: > Look, until you have any numbers in hands it's impossible to say > which one is faster. Well, it's not too hard to guess: struct dentry is 124 bytes on i386 so kmem_cache_zalloc over it is bound to be slower tha kmem_cache_alloc as most members are initialized to non-NULL. And if you're into micro-benchmarks, here's one: d_alloc takes 182.75ns d_alloc_zalloc takes 315.89ns So definitely NAK for Denis' patch. Pekka #include #include #include #include #include #include struct dentry { void *a; void *b; void *c; void *d; void *e; void *f; void *g; void *h; unsigned char d_iname[36]; }; #define NR_ITERS 1000000 #define MAX_DENTRIES (NR_ITERS*2) static unsigned long offset; static struct dentry dentries[MAX_DENTRIES]; static inline void *__kmalloc(void) { void *p = &dentries[offset++]; if (offset == MAX_DENTRIES) offset = 0; return p; } void *kmalloc(size_t sz) { return __kmalloc(); } struct dentry *d_alloc(const char *name) { struct dentry *ret = kmalloc(sizeof *ret); if (ret) { ret->a = NULL; ret->b = NULL; ret->c = NULL; ret->d = NULL; ret->e = (void *) 0xdeadbeef; ret->f = (void *) 0xdeadbeef; ret->g = (void *) 0xdeadbeef; ret->h = (void *) 0xdeadbeef; strcpy(ret->d_iname, name); } return ret; } static inline void * __memset_generic(void * s, char c,size_t count) { int d0, d1; __asm__ __volatile__( "rep\n\t" "stosb" : "=&c" (d0), "=&D" (d1) :"a" (c),"1" (s),"0" (count) :"memory"); return s; } void *kzalloc(size_t sz) { void *p = __kmalloc(); if (p) __memset_generic(p, 0, sz); return p; } struct dentry *d_alloc_zalloc(const char *name) { struct dentry *ret = kzalloc(sizeof *ret); if (ret) { ret->e = (void *) 0xdeadbeef; ret->f = (void *) 0xdeadbeef; ret->g = (void *) 0xdeadbeef; ret->h = (void *) 0xdeadbeef; strcpy(ret->d_iname, name); } return ret; } int main(int argc, char *argv[]) { struct timeval start, end; unsigned long long usec; int i; gettimeofday(&start, NULL); for (i = 0; i < NR_ITERS; i++) { struct dentry *dentry = d_alloc("root"); if (!dentry || dentry->a != NULL) abort(); } gettimeofday(&end, NULL); usec = end.tv_usec + 1000000*(end.tv_sec - start.tv_sec) - start.tv_usec; printf("d_alloc takes %0.2lfns\n", (double)usec * 1000 / NR_ITERS); gettimeofday(&start, NULL); for (i = 0; i < NR_ITERS; i++) { struct dentry *dentry = d_alloc_zalloc("root"); if (!dentry || dentry->a != NULL) abort(); } gettimeofday(&end, NULL); usec = end.tv_usec + 1000000*(end.tv_sec - start.tv_sec) - start.tv_usec; printf("d_alloc_zalloc takes %0.2lfns\n", (double)usec * 1000 / NR_ITERS); return 0; }