Re: [PATCH v5 41/50] multi-process/mig: Enable VMSD save in the Proxy object

[Jag Raman]

On 3/5/2020 7:31 AM, Dr. David Alan Gilbert wrote:
> * Jagannathan Raman (address@hidden) wrote:
> > Collect the VMSD from remote process on the source and save
> > it to the channel leading to the destination
> >
> > Signed-off-by: Elena Ufimtseva <address@hidden>
> > Signed-off-by: John G Johnson <address@hidden>
> > Signed-off-by: Jagannathan Raman <address@hidden>
> > ---
> >   v4 -> v5:
> >    - Using qemu_file_shutdown() instead of qemu_thread_cancel(). Removed patch
> >      which introduced qemu_thread_cancel()
> >
> >   hw/proxy/qemu-proxy.c         | 135 ++++++++++++++++++++++++++++++++++++++++++
> >   include/hw/proxy/qemu-proxy.h |   2 +
> >   include/io/mpqemu-link.h      |   1 +
> >   3 files changed, 138 insertions(+)
> >
> > diff --git a/hw/proxy/qemu-proxy.c b/hw/proxy/qemu-proxy.c
> > index b1b9282..19f0dbb 100644
> > --- a/hw/proxy/qemu-proxy.c
> > +++ b/hw/proxy/qemu-proxy.c
> > @@ -23,6 +23,14 @@
> >   #include "util/event_notifier-posix.c"
> >   #include "hw/boards.h"
> >   #include "include/qemu/log.h"
> > +#include "io/channel.h"
> > +#include "migration/qemu-file-types.h"
> > +#include "qapi/error.h"
> > +#include "io/channel-util.h"
> > +#include "migration/qemu-file-channel.h"
> > +#include "migration/qemu-file.h"
> > +#include "migration/migration.h"
> > +#include "migration/vmstate.h"
> >   
> >   QEMUTimer *hb_timer;
> >   static void pci_proxy_dev_realize(PCIDevice *dev, Error **errp);
> > @@ -35,6 +43,9 @@ static void broadcast_init(void);
> >   static int config_op_send(PCIProxyDev *dev, uint32_t addr, uint32_t *val, int 
> > l,
> >                             unsigned int op);
> >   
> > +#define PAGE_SIZE qemu_real_host_page_size
> > +uint8_t *mig_data;
> > +
> >   static void childsig_handler(int sig, siginfo_t *siginfo, void *ctx)
> >   {
> >       /* TODO: Add proper handler. */
> > @@ -460,6 +471,129 @@ static void pci_proxy_dev_inst_init(Object *obj)
> >       dev->mem_init = false;
> >   }
> >   
> > +typedef struct {
> > +    QEMUFile *rem;
> > +    PCIProxyDev *dev;
> > +} proxy_mig_data;
> > +
> > +static void *proxy_mig_out(void *opaque)
> > +{
> > +    proxy_mig_data *data = opaque;
> > +    PCIProxyDev *dev = data->dev;
> > +    uint8_t byte;
> > +    uint64_t data_size = PAGE_SIZE;
> > +
> > +    mig_data = g_malloc(data_size);
> > +
> > +    while (true) {
> > +        byte = qemu_get_byte(data->rem);
> > +
> > +        if (qemu_file_get_error(data->rem)) {
> > +            break;
> > +        }
> > +
> > +        mig_data[dev->migsize++] = byte;
> > +        if (dev->migsize == data_size) {
> > +            data_size += PAGE_SIZE;
> > +            mig_data = g_realloc(mig_data, data_size);
> > +        }
> > +    }
> > +
> > +    return NULL;
> > +}
> > +
> > +static int proxy_pre_save(void *opaque)
> > +{
> > +    PCIProxyDev *pdev = opaque;
> > +    proxy_mig_data *mig_data;
> > +    QEMUFile *f_remote;
> > +    MPQemuMsg msg = {0};
> > +    QemuThread thread;
> > +    Error *err = NULL;
> > +    QIOChannel *ioc;
> > +    uint64_t size;
> > +    int fd[2];
> > +
> > +    if (socketpair(AF_UNIX, SOCK_STREAM, 0, fd)) {
> > +        return -1;
> > +    }
> > +
> > +    ioc = qio_channel_new_fd(fd[0], &err);
> > +    if (err) {
> > +        error_report_err(err);
> > +        return -1;
> > +    }
> > +
> > +    qio_channel_set_name(QIO_CHANNEL(ioc), "PCIProxyDevice-mig");
> > +
> > +    f_remote = qemu_fopen_channel_input(ioc);
> > +
> > +    pdev->migsize = 0;
> > +
> > +    mig_data = g_malloc0(sizeof(proxy_mig_data));
> > +    mig_data->rem = f_remote;
>
> This feels way too complicated.   Since we know f_remote is always just
> a simple fd we're getting we don't need to use qio or qemu_file here;
> just use the fd - nice and simple.
>
> Then the code to read it can just use read(2) with a sensible size chunk
> rather than reading a byte at a time.

Hi Dave,

Upon closer inspection, we found that the migration code on the remote
(which uses QEMUFile) could sometimes set an error on the channel using
qemu_file_set_error(). We needed to use qemu_file_get_error() to catch
these errors and abort migration. So we had to stick with QEMUFile at
the receiving end as well.

> > +    mig_data->dev = pdev;
> > +
> > +    qemu_thread_create(&thread, "Proxy MIG_OUT", proxy_mig_out, mig_data,
> > +                       QEMU_THREAD_DETACHED);
>
> I'm just checking why a thread is necessary; is this because you need to
> be able to start reading before you block waiting for the remote to tell
> you the size - worrying that if you don't start reading then the remote
> might block waiting for us?

Yes, Dave. That is correct.

> > +    msg.cmd = START_MIG_OUT;
> > +    msg.bytestream = 0;
> > +    msg.num_fds = 2;
> > +    msg.fds[0] = fd[1];
> > +    msg.fds[1] = GET_REMOTE_WAIT;
> > +
> > +    mpqemu_msg_send(&msg, pdev->mpqemu_link->com);
> > +    size = wait_for_remote(msg.fds[1]);
> > +    PUT_REMOTE_WAIT(msg.fds[1]);
> > +
> > +    assert(size != ULLONG_MAX);
> > +
> > +    /*
> > +     * migsize is being update by a separate thread. Using volatile to
> > +     * instruct the compiler to fetch the value of this variable from
> > +     * memory during every read
> > +     */
> > +    while (*((volatile uint64_t *)&pdev->migsize) < size) {
> > +    }
>
> Hmm.  I suggest the following:
>
>    a) You create a shared 'size' variable;  and initialize it to
>      UINT64_MAX.
>    b) The thread uses atomic_read(shared_size) to read that value.
>    c) When you receive the size from the remote you do
>       atomic_set(&shared_size, size);
>    d) The thread does:
>       while (received_size < atomic_read(&shared_size))
>
>       so the thread will quit either on EOF or it hitting the size.
>
>    e) We pthread_join here to wait for the thread
>    f) We then check a received size to make sure it matches what we
> expect.
>
> That removes the tight loop.

Sure, will do.

> > +    qemu_file_shutdown(f_remote);
> > +
> > +    qemu_fclose(f_remote);
> > +    close(fd[1]);
> > +
> > +    return 0;
> > +}
> > +
> > +static int proxy_post_save(void *opaque)
> > +{
> > +    MigrationState *ms = migrate_get_current();
> > +    PCIProxyDev *pdev = opaque;
> > +    uint64_t pos = 0;
> > +
> > +    while (pos < pdev->migsize) {
> > +        qemu_put_byte(ms->to_dst_file, mig_data[pos]);
> > +        pos++;
> > +    }
> > +
> > +    qemu_fflush(ms->to_dst_file);
> > +
> > +    return 0;
>
> I don't think you need that.
>
> > +}
> > +
> > +const VMStateDescription vmstate_pci_proxy_device = {
> > +    .name = "PCIProxyDevice",
> > +    .version_id = 2,
> > +    .minimum_version_id = 1,
> > +    .pre_save = proxy_pre_save,
> > +    .post_save = proxy_post_save,
> > +    .fields = (VMStateField[]) {
> > +        VMSTATE_PCI_DEVICE(parent_dev, PCIProxyDev),
> > +        VMSTATE_UINT64(migsize, PCIProxyDev),
>
> I think instead you should use a VMSTATE_VBUFFER here to save
> the mig_data.
> What we should do is the post_save should g_free the buffer.
> (mig_data should be a field in proxy).

We noticed that VMSTATE_BUFFER requires that the buffer be part of a
"struct" and that the buffer is an array. Since the buffer we're using
is neither an array nor part of a "struct", we decided to go with
writing the buffer directly to the migration stream in proxy_post_save()
instead of using VMSTATE_BUFFER.

I think we should nevertheless g_free this buffer in post_save like you
pointed out.

Thank you!
--
Jag

> Dave
>
>
> > +        VMSTATE_END_OF_LIST()
> > +    }
> > +};
> > +
> >   static void pci_proxy_dev_class_init(ObjectClass *klass, void *data)
> >   {
> >       PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
> > @@ -471,6 +605,7 @@ static void pci_proxy_dev_class_init(ObjectClass *klass, 
> > void *data)
> >       k->config_write = pci_proxy_write_config;
> >   
> >       dc->reset = proxy_device_reset;
> > +    dc->vmsd = &vmstate_pci_proxy_device;
> >   }
> >   
> >   static const TypeInfo pci_proxy_dev_type_info = {
> > diff --git a/include/hw/proxy/qemu-proxy.h b/include/hw/proxy/qemu-proxy.h
> > index 5de8129..537c227 100644
> > --- a/include/hw/proxy/qemu-proxy.h
> > +++ b/include/hw/proxy/qemu-proxy.h
> > @@ -75,6 +75,8 @@ struct PCIProxyDev {
> >                           bool need_spawn, Error **errp);
> >   
> >       ProxyMemoryRegion region[PCI_NUM_REGIONS];
> > +
> > +    uint64_t migsize;
> >   };
> >   
> >   typedef struct PCIProxyDevClass {
> > diff --git a/include/io/mpqemu-link.h b/include/io/mpqemu-link.h
> > index d2234ca..b42c003 100644
> > --- a/include/io/mpqemu-link.h
> > +++ b/include/io/mpqemu-link.h
> > @@ -63,6 +63,7 @@ typedef enum {
> >       PROXY_PING,
> >       MMIO_RETURN,
> >       DEVICE_RESET,
> > +    START_MIG_OUT,
> >       MAX,
> >   } mpqemu_cmd_t;
> >   
> > --
> > 1.8.3.1
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK