view src/core/ngx_cpuinfo.c @ 5569:462ae7eedc68 stable-1.4

Fixed TCP_DEFER_ACCEPT handling (ticket #353). Backed out 05a56ebb084a, as it turns out that kernel can return connections without any delay if syncookies are used. This basically means we can't assume anything about connections returned with deferred accept set. To solve original problem the 05a56ebb084a tried to solve, i.e. to don't wait longer than needed if a connection was accepted after deferred accept timeout, this patch changes a timeout set with setsockopt(TCP_DEFER_ACCEPT) to 1 second, unconditionally. This is believed to be enough for speed improvements, and doesn't imply major changes to timeouts used. Note that before 2.6.32 connections were dropped after a timeout. Though it is believed that 1s is still appropriate for kernels before 2.6.32, as previously tcp_synack_retries controlled the actual timeout and 1s results in more than 1 minute actual timeout by default.
author Maxim Dounin <mdounin@mdounin.ru>
date Tue, 28 Jan 2014 15:40:46 +0400
parents d620f497c50f
children
line wrap: on
line source


/*
 * Copyright (C) Igor Sysoev
 * Copyright (C) Nginx, Inc.
 */


#include <ngx_config.h>
#include <ngx_core.h>


#if (( __i386__ || __amd64__ ) && ( __GNUC__ || __INTEL_COMPILER ))


static ngx_inline void ngx_cpuid(uint32_t i, uint32_t *buf);


#if ( __i386__ )

static ngx_inline void
ngx_cpuid(uint32_t i, uint32_t *buf)
{

    /*
     * we could not use %ebx as output parameter if gcc builds PIC,
     * and we could not save %ebx on stack, because %esp is used,
     * when the -fomit-frame-pointer optimization is specified.
     */

    __asm__ (

    "    mov    %%ebx, %%esi;  "

    "    cpuid;                "
    "    mov    %%eax, (%1);   "
    "    mov    %%ebx, 4(%1);  "
    "    mov    %%edx, 8(%1);  "
    "    mov    %%ecx, 12(%1); "

    "    mov    %%esi, %%ebx;  "

    : : "a" (i), "D" (buf) : "ecx", "edx", "esi", "memory" );
}


#else /* __amd64__ */


static ngx_inline void
ngx_cpuid(uint32_t i, uint32_t *buf)
{
    uint32_t  eax, ebx, ecx, edx;

    __asm__ (

        "cpuid"

    : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx) : "a" (i) );

    buf[0] = eax;
    buf[1] = ebx;
    buf[2] = edx;
    buf[3] = ecx;
}


#endif


/* auto detect the L2 cache line size of modern and widespread CPUs */

void
ngx_cpuinfo(void)
{
    u_char    *vendor;
    uint32_t   vbuf[5], cpu[4], model;

    vbuf[0] = 0;
    vbuf[1] = 0;
    vbuf[2] = 0;
    vbuf[3] = 0;
    vbuf[4] = 0;

    ngx_cpuid(0, vbuf);

    vendor = (u_char *) &vbuf[1];

    if (vbuf[0] == 0) {
        return;
    }

    ngx_cpuid(1, cpu);

    if (ngx_strcmp(vendor, "GenuineIntel") == 0) {

        switch ((cpu[0] & 0xf00) >> 8) {

        /* Pentium */
        case 5:
            ngx_cacheline_size = 32;
            break;

        /* Pentium Pro, II, III */
        case 6:
            ngx_cacheline_size = 32;

            model = ((cpu[0] & 0xf0000) >> 8) | (cpu[0] & 0xf0);

            if (model >= 0xd0) {
                /* Intel Core, Core 2, Atom */
                ngx_cacheline_size = 64;
            }

            break;

        /*
         * Pentium 4, although its cache line size is 64 bytes,
         * it prefetches up to two cache lines during memory read
         */
        case 15:
            ngx_cacheline_size = 128;
            break;
        }

    } else if (ngx_strcmp(vendor, "AuthenticAMD") == 0) {
        ngx_cacheline_size = 64;
    }
}

#else


void
ngx_cpuinfo(void)
{
}


#endif