Commit c8090972 authored by Peter Maydell's avatar Peter Maydell
Browse files

Merge remote-tracking branch 'remotes/marcel/tags/rdma-pull-request' into staging 



RDMA queue

# gpg: Signature made Sat 18 Aug 2018 16:01:46 BST
# gpg:                using RSA key 36D4C0F0CF2FE46D
# gpg: Good signature from "Marcel Apfelbaum <marcel.apfelbaum@zoho.com>"
# gpg:                 aka "Marcel Apfelbaum <marcel@redhat.com>"
# gpg:                 aka "Marcel Apfelbaum <marcel.apfelbaum@gmail.com>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: B1C6 3A57 F92E 08F2 640F  31F5 36D4 C0F0 CF2F E46D

* remotes/marcel/tags/rdma-pull-request:
  config: split PVRDMA from RDMA
  hw/pvrdma: remove not needed include
  hw/rdma: Add reference to pci_dev in backend_dev
  hw/rdma: Bugfix - Support non-aligned buffers
  hw/rdma: Print backend QP number in hex format
  hw/rdma: Cosmetic change - move to generic function
  hw/pvrdma: Cosmetic change - indent right
  hw/rdma: Reorder resource cleanup
  hw/rdma: Do not allocate memory for non-dma MR
  hw/rdma: Delete useless structure RdmaRmUserMR
  hw/pvrdma: Make default pkey 0xFFFF
  hw/pvrdma: Clean CQE before use
  hw/rdma: Modify debug macros
  hw/pvrdma: Bugfix - provide the correct attr_mask to query_qp
  hw/rdma: Make distinction between device init and start modes

Signed-off-by: default avatarPeter Maydell <peter.maydell@linaro.org>
parents 62c34848 21ab34c9

configure

+54 −1
Original line number Diff line number Diff line
@@ -375,6 +375,7 @@ hax="no" 
hvf="no"

whpx="no"

rdma=""

pvrdma=""

gprof="no"

debug_tcg="no"

debug="no"
@@ -1363,6 +1364,10 @@ for opt do 
  ;;

  --disable-rdma) rdma="no"

  ;;

  --enable-pvrdma) pvrdma="yes"

  ;;

  --disable-pvrdma) pvrdma="no"

  ;;

  --with-gtkabi=*) gtkabi="$optarg"

  ;;

  --disable-vte) vte="no"
@@ -1669,7 +1674,8 @@ disabled with --disable-FEATURE, default is enabled if available: 
  hax             HAX acceleration support

  hvf             Hypervisor.framework acceleration support

  whpx            Windows Hypervisor Platform acceleration support

  rdma            Enable RDMA-based migration and PVRDMA support

  rdma            Enable RDMA-based migration

  pvrdma          Enable PVRDMA support

  vde             support for vde network

  netmap          support for netmap network

  linux-aio       Linux AIO support
@@ -3064,6 +3070,48 @@ EOF 
  fi

fi



##########################################

# PVRDMA detection



cat > $TMPC <<EOF &&

#include <sys/mman.h>



int

main(void)

{

    char buf = 0;

    void *addr = &buf;

    addr = mremap(addr, 0, 1, MREMAP_MAYMOVE | MREMAP_FIXED);



    return 0;

}

EOF



if test "$rdma" = "yes" ; then

    case "$pvrdma" in

    "")

        if compile_prog "" ""; then

            pvrdma="yes"

        else

            pvrdma="no"

        fi

        ;;

    "yes")

        if ! compile_prog "" ""; then

            error_exit "PVRDMA is not supported since mremap is not implemented"

        fi

        pvrdma="yes"

        ;;

    "no")

        pvrdma="no"

        ;;

    esac

else

    if test "$pvrdma" = "yes" ; then

        error_exit "PVRDMA requires rdma suppport"

    fi

    pvrdma="no"

fi



##########################################

# VNC SASL detection
@@ -5952,6 +6000,7 @@ if test "$tcg" = "yes" ; then 
fi

echo "malloc trim support $malloc_trim"

echo "RDMA support      $rdma"

echo "PVRDMA support    $pvrdma"

echo "fdt support       $fdt"

echo "membarrier        $membarrier"

echo "preadv support    $preadv"
@@ -6708,6 +6757,10 @@ if test "$rdma" = "yes" ; then 
  echo "RDMA_LIBS=$rdma_libs" >> $config_host_mak

fi



if test "$pvrdma" = "yes" ; then

  echo "CONFIG_PVRDMA=y" >> $config_host_mak

fi



if test "$have_rtnetlink" = "yes" ; then

  echo "CONFIG_RTNETLINK=y" >> $config_host_mak

fi

hw/rdma/Makefile.objs

+1 −1
Original line number Diff line number Diff line 
ifeq ($(CONFIG_RDMA),y)

ifeq ($(CONFIG_PVRDMA),y)

obj-$(CONFIG_PCI) += rdma_utils.o rdma_backend.o rdma_rm.o

obj-$(CONFIG_PCI) += vmw/pvrdma_dev_ring.o vmw/pvrdma_cmd.o \

                     vmw/pvrdma_qp_ops.o vmw/pvrdma_main.o

hw/rdma/rdma_backend.c

+83 −22
Original line number Diff line number Diff line
@@ -35,6 +35,7 @@ 
#define VENDOR_ERR_MR_SMALL         0x208



#define THR_NAME_LEN 16

#define THR_POLL_TO  5000



typedef struct BackendCtx {

    uint64_t req_id;
@@ -91,10 +92,30 @@ static void *comp_handler_thread(void *arg) 
    int rc;

    struct ibv_cq *ev_cq;

    void *ev_ctx;

    int flags;

    GPollFD pfds[1];



    /* Change to non-blocking mode */

    flags = fcntl(backend_dev->channel->fd, F_GETFL);

    rc = fcntl(backend_dev->channel->fd, F_SETFL, flags | O_NONBLOCK);

    if (rc < 0) {

        pr_dbg("Fail to change to non-blocking mode\n");

        return NULL;

    }



    pr_dbg("Starting\n");



    pfds[0].fd = backend_dev->channel->fd;

    pfds[0].events = G_IO_IN | G_IO_HUP | G_IO_ERR;



    backend_dev->comp_thread.is_running = true;



    while (backend_dev->comp_thread.run) {

        do {

            rc = qemu_poll_ns(pfds, 1, THR_POLL_TO * (int64_t)SCALE_MS);

        } while (!rc && backend_dev->comp_thread.run);



        if (backend_dev->comp_thread.run) {

            pr_dbg("Waiting for completion on channel %p\n", backend_dev->channel);

            rc = ibv_get_cq_event(backend_dev->channel, &ev_cq, &ev_ctx);

            pr_dbg("ibv_get_cq_event=%d\n", rc);
@@ -112,14 +133,41 @@ static void *comp_handler_thread(void *arg) 


            ibv_ack_cq_events(ev_cq, 1);

        }

    }



    pr_dbg("Going down\n");



    /* TODO: Post cqe for all remaining buffs that were posted */



    backend_dev->comp_thread.is_running = false;



    qemu_thread_exit(0);



    return NULL;

}



static void stop_backend_thread(RdmaBackendThread *thread)

{

    thread->run = false;

    while (thread->is_running) {

        pr_dbg("Waiting for thread to complete\n");

        sleep(THR_POLL_TO / SCALE_US / 2);

    }

}



static void start_comp_thread(RdmaBackendDev *backend_dev)

{

    char thread_name[THR_NAME_LEN] = {0};



    stop_backend_thread(&backend_dev->comp_thread);



    snprintf(thread_name, sizeof(thread_name), "rdma_comp_%s",

             ibv_get_device_name(backend_dev->ib_dev));

    backend_dev->comp_thread.run = true;

    qemu_thread_create(&backend_dev->comp_thread.thread, thread_name,

                       comp_handler_thread, backend_dev, QEMU_THREAD_DETACHED);

}



void rdma_backend_register_comp_handler(void (*handler)(int status,

                                        unsigned int vendor_err, void *ctx))

{
@@ -223,8 +271,7 @@ static int build_host_sge_array(RdmaDeviceResources *rdma_dev_res, 
            return VENDOR_ERR_INVLKEY | ssge[ssge_idx].lkey;

        }



        dsge->addr = (uintptr_t)mr->user_mr.host_virt + ssge[ssge_idx].addr -

                     mr->user_mr.guest_start;

        dsge->addr = (uintptr_t)mr->virt + ssge[ssge_idx].addr - mr->start;

        dsge->length = ssge[ssge_idx].length;

        dsge->lkey = rdma_backend_mr_lkey(&mr->backend_mr);
@@ -697,7 +744,7 @@ static int init_device_caps(RdmaBackendDev *backend_dev, 
    return 0;

}



int rdma_backend_init(RdmaBackendDev *backend_dev,

int rdma_backend_init(RdmaBackendDev *backend_dev, PCIDevice *pdev,

                      RdmaDeviceResources *rdma_dev_res,

                      const char *backend_device_name, uint8_t port_num,

                      uint8_t backend_gid_idx, struct ibv_device_attr *dev_attr,
@@ -706,10 +753,13 @@ int rdma_backend_init(RdmaBackendDev *backend_dev, 
    int i;

    int ret = 0;

    int num_ibv_devices;

    char thread_name[THR_NAME_LEN] = {0};

    struct ibv_device **dev_list;

    struct ibv_port_attr port_attr;



    memset(backend_dev, 0, sizeof(*backend_dev));



    backend_dev->dev = pdev;



    backend_dev->backend_gid_idx = backend_gid_idx;

    backend_dev->port_num = port_num;

    backend_dev->rdma_dev_res = rdma_dev_res;
@@ -800,11 +850,8 @@ int rdma_backend_init(RdmaBackendDev *backend_dev, 
    pr_dbg("interface_id=0x%" PRIx64 "\n",

           be64_to_cpu(backend_dev->gid.global.interface_id));



    snprintf(thread_name, sizeof(thread_name), "rdma_comp_%s",

             ibv_get_device_name(backend_dev->ib_dev));

    backend_dev->comp_thread.run = true;

    qemu_thread_create(&backend_dev->comp_thread.thread, thread_name,

                       comp_handler_thread, backend_dev, QEMU_THREAD_DETACHED);

    backend_dev->comp_thread.run = false;

    backend_dev->comp_thread.is_running = false;



    ah_cache_init();
@@ -823,8 +870,22 @@ out: 
    return ret;

}





void rdma_backend_start(RdmaBackendDev *backend_dev)

{

    pr_dbg("Starting rdma_backend\n");

    start_comp_thread(backend_dev);

}



void rdma_backend_stop(RdmaBackendDev *backend_dev)

{

    pr_dbg("Stopping rdma_backend\n");

    stop_backend_thread(&backend_dev->comp_thread);

}



void rdma_backend_fini(RdmaBackendDev *backend_dev)

{

    rdma_backend_stop(backend_dev);

    g_hash_table_destroy(ah_hash);

    ibv_destroy_comp_channel(backend_dev->channel);

    ibv_close_device(backend_dev->context);

hw/rdma/rdma_backend.h

+3 −1
Original line number Diff line number Diff line
@@ -46,12 +46,14 @@ static inline uint32_t rdma_backend_mr_rkey(const RdmaBackendMR *mr) 
    return mr->ibmr ? mr->ibmr->rkey : 0;

}



int rdma_backend_init(RdmaBackendDev *backend_dev,

int rdma_backend_init(RdmaBackendDev *backend_dev, PCIDevice *pdev,

                      RdmaDeviceResources *rdma_dev_res,

                      const char *backend_device_name, uint8_t port_num,

                      uint8_t backend_gid_idx, struct ibv_device_attr *dev_attr,

                      Error **errp);

void rdma_backend_fini(RdmaBackendDev *backend_dev);

void rdma_backend_start(RdmaBackendDev *backend_dev);

void rdma_backend_stop(RdmaBackendDev *backend_dev);

void rdma_backend_register_comp_handler(void (*handler)(int status,

                                        unsigned int vendor_err, void *ctx));

void rdma_backend_unregister_comp_handler(void);

hw/rdma/rdma_backend_defs.h

+2 −1
Original line number Diff line number Diff line
@@ -24,7 +24,8 @@ typedef struct RdmaDeviceResources RdmaDeviceResources; 
typedef struct RdmaBackendThread {

    QemuThread thread;

    QemuMutex mutex;

    bool run;

    bool run; /* Set by thread manager to let thread know it should exit */

    bool is_running; /* Set by the thread to report its status */

} RdmaBackendThread;



typedef struct RdmaBackendDev {