libldom/sparc/ldom.c

	ldom.c revision a298b85fd4d5149fdcc1bae48f9baab83a0390c9
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#pragma ident   "%Z%%M% %I% %E% SMI"

#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <string.h>
#include <fcntl.h>
#include <pthread.h>
#include <errno.h>
#include <libnvpair.h>

#include <sys/processor.h>
#include <sys/stat.h>
#include <sys/mdesc.h>
#include <sys/param.h>
#include <sys/systeminfo.h>
#include <sys/mem.h>
#include <sys/bl.h>
#include <sys/fm/protocol.h>
#include <fm/fmd_fmri.h>

#include "ldom.h"
#include "ldmsvcs_utils.h"


static ssize_t
get_local_core_md(ldom_hdl_t *lhp, uint64_t **buf)
{
    int fh;
    size_t size;
    uint64_t *bufp;

    if ((fh = open("/devices/pseudo/mdesc@0:mdesc", O_RDONLY, 0)) < 0)
        return (-1);

    if (ioctl(fh, MDESCIOCGSZ, &size) < 0) {
        (void) close(fh);
        return (-1);
    }

    bufp = (uint64_t *)lhp->allocp(size);

    if (read(fh, bufp, size) < 0) {
        lhp->freep(bufp, size);
        (void) close(fh);
        return (-1);
    }
    (void) close(fh);

    *buf = bufp;

    return ((ssize_t)size);
}


static int
ldom_getinfo(struct ldom_hdl *lhp)
{
    static pthread_mutex_t mt = PTHREAD_MUTEX_INITIALIZER;
    static pthread_cond_t cv = PTHREAD_COND_INITIALIZER;
    static int major_version = -1;
    static int service_ldom = -1;
    static int busy_init = 0;

    int ier, rc = 0;

    (void) pthread_mutex_lock(&mt);

    while (busy_init == 1)
        (void) pthread_cond_wait(&cv, &mt);

    if (major_version != -1 && service_ldom != -1) {
        lhp->major_version = major_version;
        lhp->service_ldom = service_ldom;
        (void) pthread_mutex_unlock(&mt);
        return (0);
    }

    /*
     * get to this point if major_version and service_ldom have not yet
     * been determined
     */
    busy_init = 1;
    (void) pthread_mutex_unlock(&mt);

    /*
     * set defaults which correspond to the case of "LDOMS not
     * available".  note that these can (and will) also apply to
     * non-sun4v machines.
     */
    major_version = 0;
    service_ldom = 1;

    /* figure out version */
    if ((ier = ldmsvcs_check_channel()) == 0) {
        /*
         * get into this block if vldc exists.  LDOMS is available
         * and we are on the service LDOM.
         */
        major_version = 1;
        service_ldom = 1;
    } else if (ier == 1) {
        /*
         * get into this block if vldc does not exist
         *
         * if we do not get into the following if() block [i.e.,
         * if (bufsiz <= 0)] then we are on a non-sun4v machine.
         */
        uint64_t *bufp;
        ssize_t bufsiz;

        if ((bufsiz = get_local_core_md(lhp, &bufp)) > 0) {
            md_t *mdp;

            if ((mdp = md_init_intern(bufp, lhp->allocp,
                            lhp->freep)) != NULL) {
                mde_cookie_t *listp;
                uint64_t dval;
                int num_nodes;

                num_nodes = md_node_count(mdp);
                listp = lhp->allocp(sizeof (mde_cookie_t) *
                            num_nodes);

                /*
                 * if we do not enter the following if block,
                 * we conclude that LDOMS is not available
                 */
                if (md_scan_dag(mdp, MDE_INVAL_ELEM_COOKIE,
                    md_find_name(mdp, "platform"),
                    md_find_name(mdp, "fwd"),
                    listp) > 0 &&
                    md_get_prop_val(mdp, listp[0],
                    "domaining-enabled", &dval) >= 0 &&
                    dval == 1) {
                    /*
                     * LDOMS is available.  an earlier
                     * block detected the situation of
                     * being on a service LDOM, so
                     * we get to this point only if we
                     * are not on a service LDOM.
                     */
                    major_version = 1;
                    service_ldom = 0;
                }

                lhp->freep(listp, sizeof (mde_cookie_t) *
                        num_nodes);
                (void) md_fini(mdp);
            }

            lhp->freep(bufp, bufsiz);
        }
    } else {
        rc = 1;
    }

    (void) pthread_mutex_lock(&mt);
    lhp->major_version = major_version;
    lhp->service_ldom = service_ldom;
    busy_init = 0;
    (void) pthread_mutex_unlock(&mt);

    (void) pthread_cond_broadcast(&cv);

    return (rc);
}


/*
 * search the machine description for a "pid" entry (physical cpuid) and
 * return the corresponding "id" entry (virtual cpuid)
 */
static processorid_t
cpu_phys2virt(ldom_hdl_t *lhp, uint32_t cpuid)
{
    char isa[MAXNAMELEN];
    md_t *mdp;
    mde_cookie_t *listp;
    ssize_t bufsize;
    processorid_t vid;
    uint64_t *bufp;
    uint64_t pval;
    int num_nodes, ncpus, i;

    (void) sysinfo(SI_ARCHITECTURE, isa, MAXNAMELEN);

    if (strcmp(isa, "sun4v") != 0)
        return ((processorid_t)cpuid);

    /*
     * convert the physical cpuid to a virtual cpuid
     */
    if ((bufsize = ldom_get_core_md(lhp, &bufp)) < 1)
        return (-1);

    if ((mdp = md_init_intern(bufp, lhp->allocp, lhp->freep)) == NULL ||
        (num_nodes = md_node_count(mdp)) < 1) {
        lhp->freep(bufp, bufsize);
        return (-1);
    }

    listp = (mde_cookie_t *)lhp->allocp(sizeof (mde_cookie_t) * num_nodes);
    ncpus = md_scan_dag(mdp, MDE_INVAL_ELEM_COOKIE,
                md_find_name(mdp, "cpu"),
                md_find_name(mdp, "fwd"), listp);

    vid = -1;
    for (i = 0; i < ncpus; i++) {
        if (md_get_prop_val(mdp, listp[i], "pid", &pval) >= 0 &&
            pval == (uint64_t)cpuid) {
            if (md_get_prop_val(mdp, listp[i], "id", &pval) >= 0)
                vid = (processorid_t)pval;

            break;
        }
    }

    lhp->freep(listp, sizeof (mde_cookie_t) * num_nodes);
    (void) md_fini(mdp);
    lhp->freep(bufp, bufsize);

    return (vid);
}

/*
 * if checking for status of a retired page:
 *   0 - page is retired
 *   EAGAIN - page is scheduled for retirement
 *   EIO - page not scheduled for retirement
 *   EINVAL - error
 *
 * if retiring a page:
 *   0 - success in retiring page
 *   EIO - page is already retired
 *   EAGAIN - page is scheduled for retirement
 *   EINVAL - error
 *
 * the original decoder for ioctl() return values is
 * http://fma.eng/documents/engineering/cpumem/page_retire_api.txt
 */
static int
os_mem_page_retire(ldom_hdl_t *lhp, int cmd, nvlist_t *nvl)
{
    mem_page_t mpage;
    char *fmribuf;
    size_t fmrisz;
    int fd, rc, err;

    if (cmd != MEM_PAGE_RETIRE && cmd != MEM_PAGE_FMRI_RETIRE &&
        cmd != MEM_PAGE_ISRETIRED && cmd != MEM_PAGE_FMRI_ISRETIRED)
            return (EINVAL);

    if ((fd = open("/dev/mem", O_RDONLY)) < 0)
        return (EINVAL);

    if ((errno = nvlist_size(nvl, &fmrisz, NV_ENCODE_NATIVE)) != 0 ||
        fmrisz > MEM_FMRI_MAX_BUFSIZE ||
        (fmribuf = lhp->allocp(fmrisz)) == NULL) {
        (void) close(fd);
        return (EINVAL);
    }

    if ((errno = nvlist_pack(nvl, &fmribuf, &fmrisz,
                    NV_ENCODE_NATIVE, 0)) != 0) {
        lhp->freep(fmribuf, fmrisz);
        (void) close(fd);
        return (EINVAL);
    }

    mpage.m_fmri = fmribuf;
    mpage.m_fmrisz = fmrisz;

    rc = ioctl(fd, cmd, &mpage);
    err = errno;

    lhp->freep(fmribuf, fmrisz);
    (void) close(fd);

    if (rc < 0) {
        rc = err;
    }

    return (rc);
}

int
ldom_fmri_status(ldom_hdl_t *lhp, nvlist_t *nvl)
{
    char *name;
    int ret;

    if (nvlist_lookup_string(nvl, FM_FMRI_SCHEME, &name) != 0)
        return (EINVAL);

    switch (ldom_major_version(lhp)) {
    case 0:
        /*
         * version == 0 means LDOMS support is not available
         */
        if (strcmp(name, FM_FMRI_SCHEME_CPU) == 0) {
            processorid_t vid;
            uint32_t cpuid;

            if (nvlist_lookup_uint32(nvl, FM_FMRI_CPU_ID,
                            &cpuid) == 0 &&
                (vid = cpu_phys2virt(lhp, cpuid)) != -1)
                return (p_online(vid, P_STATUS));
        } else if (strcmp(name, FM_FMRI_SCHEME_MEM) == 0) {
            return (os_mem_page_retire(lhp,
                        MEM_PAGE_FMRI_ISRETIRED, nvl));
        }

        return (EINVAL);
        /*NOTREACHED*/
        break;
    case 1:
        /* LDOMS 1.0 */
        if (strcmp(name, FM_FMRI_SCHEME_CPU) == 0) {
            uint32_t cpuid;

            if (nvlist_lookup_uint32(nvl, FM_FMRI_CPU_ID,
                        &cpuid) == 0)
                ret = ldmsvcs_cpu_req_status(lhp, cpuid);
        } else if (strcmp(name, FM_FMRI_SCHEME_MEM) == 0) {
            uint64_t pa;

            if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_PHYSADDR,
                        &pa) == 0)
                ret = ldmsvcs_mem_req_status(lhp, pa);
            else
                ret = EINVAL;
        } else {
            ret = ENOTSUP;
        }
        return (ret);

        /*NOTREACHED*/
        break;
    default:
        break;
    }

    return (ENOTSUP);
}


int
ldom_fmri_retire(ldom_hdl_t *lhp, nvlist_t *nvl)
{
    char *name;
    int ret;

    if (nvlist_lookup_string(nvl, FM_FMRI_SCHEME, &name) != 0)
        return (EINVAL);

    switch (ldom_major_version(lhp)) {
    case 0:
        /*
         * version == 0 means LDOMS support is not available
         */
        if (strcmp(name, FM_FMRI_SCHEME_CPU) == 0) {
            processorid_t vid;
            uint32_t cpuid;

            if (nvlist_lookup_uint32(nvl, FM_FMRI_CPU_ID,
                            &cpuid) == 0 &&
                (vid = cpu_phys2virt(lhp, cpuid)) != -1)
                return (p_online(vid, P_FAULTED));
        } else if (strcmp(name, FM_FMRI_SCHEME_MEM) == 0) {
            return (os_mem_page_retire(lhp,
                        MEM_PAGE_FMRI_RETIRE, nvl));
        }

        return (EINVAL);
        /*NOTREACHED*/
        break;
    case 1:
        /* LDOMS 1.0 */
        if (strcmp(name, FM_FMRI_SCHEME_CPU) == 0) {
            uint32_t cpuid;

            if (nvlist_lookup_uint32(nvl, FM_FMRI_CPU_ID,
                        &cpuid) == 0)
                ret = ldmsvcs_cpu_req_offline(lhp, cpuid);
        } else if (strcmp(name, FM_FMRI_SCHEME_MEM) == 0) {
            uint64_t pa;

            if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_PHYSADDR,
                        &pa) == 0)
                ret = ldmsvcs_mem_req_retire(lhp, pa);
            else
                ret = EINVAL;
        } else {
            ret = ENOTSUP;
        }
        return (ret);

        /*NOTREACHED*/
        break;
    default:
        break;
    }

    return (ENOTSUP);
}


/*
 * blacklist cpus in a non-LDOMS environment
 */
int
ldom_fmri_blacklist(ldom_hdl_t *lhp, nvlist_t *nvl)
{
    char *name;

    if (ldom_major_version(lhp) != 0)
        return (0);

    if (nvlist_lookup_string(nvl, FM_FMRI_SCHEME, &name) != 0)
        return (EINVAL);

    if (strcmp(name, FM_FMRI_SCHEME_CPU) == 0) {
        bl_req_t blr;
        char *class;
        int fd, rc, err;

        if ((nvlist_lookup_string(nvl, FM_CLASS, &class) != 0) ||
            (class == NULL) || (*class == '\0'))
            return (EINVAL);

        if ((fd = open("/dev/bl", O_RDONLY)) < 0)
            return (EIO);

        if (nvlist_size(nvl, &blr.bl_fmrisz, NV_ENCODE_NATIVE) != 0 ||
            blr.bl_fmrisz == 0 ||
            (blr.bl_fmri = (caddr_t)lhp->allocp(blr.bl_fmrisz)) ==
            NULL) {
            (void) close(fd);
            return (EINVAL);
        }

        blr.bl_class = class;

        rc = ioctl(fd, BLIOC_INSERT, &blr);
        err = errno;

        lhp->freep((void *)&blr.bl_fmri, blr.bl_fmrisz);
        (void) close(fd);

        if (rc < 0 && err != ENOTSUP) {
            errno = err;
            return (-1);
        }
    }

    return (0);
}


ssize_t
ldom_get_core_md(ldom_hdl_t *lhp, uint64_t **buf)
{
    switch (ldom_major_version(lhp)) {
    case 0:
        return (get_local_core_md(lhp, buf));
        /*NOTREACHED*/
        break;
    case 1:
        /* LDOMS 1.0 */
        if (ldom_on_service(lhp) == 1)
            return (ldmsvcs_get_core_md(lhp, buf));
        else
            return (get_local_core_md(lhp, buf));

        /*NOTREACHED*/
        break;
    default:
        *buf = NULL;
        break;
    }

    return (-1);
}


/*
 * version 0 means no LDOMS
 */
int
ldom_major_version(ldom_hdl_t *lhp)
{
    if (lhp == NULL)
        return (-1);

    if (ldom_getinfo(lhp) == 0)
        return (lhp->major_version);
    else
        return (0);
}

/*
 * in the absence of ldoms we are on a single OS instance which is the
 * equivalent of the service ldom
 */
int
ldom_on_service(ldom_hdl_t *lhp)
{
    if (lhp == NULL)
        return (-1);

    if (ldom_getinfo(lhp) == 0)
        return (lhp->service_ldom);
    else
        return (1);
}


ldom_hdl_t *
ldom_init(void *(*allocp)(size_t size),
    void (*freep)(void *addr, size_t size))
{
    struct ldom_hdl *lhp;

    if ((lhp = allocp(sizeof (struct ldom_hdl))) == NULL)
        return (NULL);

    lhp->major_version = -1;    /* version not yet determined */
    lhp->allocp = allocp;
    lhp->freep = freep;

    ldmsvcs_init(lhp);

    return (lhp);
}


void
ldom_fini(ldom_hdl_t *lhp)
{
    if (lhp == NULL)
        return;

    ldmsvcs_fini(lhp);
    lhp->freep(lhp, sizeof (struct ldom_hdl));
}

/* end file */