view src/core/ngx_cpuinfo.c @ 8776:ec2e6893caaa

Simplified sendfile(SF_NODISKIO) usage. Starting with FreeBSD 11, there is no need to use AIO operations to preload data into cache for sendfile(SF_NODISKIO) to work. Instead, sendfile() handles non-blocking loading data from disk by itself. It still can, however, return EBUSY if a page is already being loaded (for example, by a different process). If this happens, we now post an event for the next event loop iteration, so sendfile() is retried "after a short period", as manpage recommends. The limit of the number of EBUSY tolerated without any progress is preserved, but now it does not result in an alert, since on an idle system event loop iteration might be very short and EBUSY can happen many times in a row. Instead, SF_NODISKIO is simply disabled for one call once the limit is reached. With this change, sendfile(SF_NODISKIO) is now used automatically as long as sendfile() is enabled, and no longer requires "aio on;".
author Maxim Dounin <mdounin@mdounin.ru>
date Mon, 27 Dec 2021 19:48:33 +0300
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