Re: [Qemu-devel] [PATCH v5 40/45] Postcopy; Handle userfault requests

[Dr. David Alan Gilbert]

* David Gibson (address@hidden) wrote:
> On Wed, Feb 25, 2015 at 04:52:03PM +0000, Dr. David Alan Gilbert (git) wrote:
> > From: "Dr. David Alan Gilbert" <address@hidden>
> > 
> > userfaultfd is a Linux syscall that gives an fd that receives a stream
> > of notifications of accesses to pages registered with it and allows
> > the program to acknowledge those stalls and tell the accessing
> > thread to carry on.
> > 
> > Signed-off-by: Dr. David Alan Gilbert <address@hidden>
> > ---
> >  include/migration/migration.h |   4 +
> >  migration/postcopy-ram.c      | 217 
> > ++++++++++++++++++++++++++++++++++++++++--
> >  trace-events                  |  12 +++
> >  3 files changed, 223 insertions(+), 10 deletions(-)
> > 
> > diff --git a/include/migration/migration.h b/include/migration/migration.h
> > index 139bb1b..cec064f 100644
> > --- a/include/migration/migration.h
> > +++ b/include/migration/migration.h
> > @@ -86,11 +86,15 @@ struct MigrationIncomingState {
> >  
> >      PostcopyState postcopy_state;
> >  
> > +    bool           have_fault_thread;
> >      QemuThread     fault_thread;
> >      QemuSemaphore  fault_thread_sem;
> >  
> >      /* For the kernel to send us notifications */
> >      int            userfault_fd;
> > +    /* To tell the fault_thread to quit */
> > +    int            userfault_quit_fd;
> > +
> >      QEMUFile *return_path;
> >      QemuMutex      rp_mutex;    /* We send replies from multiple threads */
> >      PostcopyPMI    postcopy_pmi;
> > diff --git a/migration/postcopy-ram.c b/migration/postcopy-ram.c
> > index 86fa5a0..abc039e 100644
> > --- a/migration/postcopy-ram.c
> > +++ b/migration/postcopy-ram.c
> > @@ -47,6 +47,8 @@ struct PostcopyDiscardState {
> >   */
> >  #if defined(__linux__)
> >  
> > +#include <poll.h>
> > +#include <sys/eventfd.h>
> >  #include <sys/mman.h>
> >  #include <sys/ioctl.h>
> >  #include <sys/types.h>
> > @@ -264,7 +266,7 @@ void postcopy_pmi_dump(MigrationIncomingState *mis)
> >  void postcopy_hook_early_receive(MigrationIncomingState *mis,
> >                                   size_t bitmap_index)
> >  {
> > -    if (mis->postcopy_state == POSTCOPY_INCOMING_ADVISE) {
> > +    if (postcopy_state_get(mis) == POSTCOPY_INCOMING_ADVISE) {
> 
> It kind of looks like that's a fix which should be folded into an
> earlier patch.

Thanks; gone.

> 
> >          /*
> >           * If we're in precopy-advise mode we need to track received pages 
> > even
> >           * though we don't need to place pages atomically yet.
> > @@ -489,15 +491,40 @@ int postcopy_ram_incoming_init(MigrationIncomingState 
> > *mis, size_t ram_pages)
> >   */
> >  int postcopy_ram_incoming_cleanup(MigrationIncomingState *mis)
> >  {
> > -    /* TODO: Join the fault thread once we're sure it will exit */
> > -    if (qemu_ram_foreach_block(cleanup_area, mis)) {
> > -        return -1;
> > +    trace_postcopy_ram_incoming_cleanup_entry();
> > +
> > +    if (mis->have_fault_thread) {
> > +        uint64_t tmp64;
> > +
> > +        if (qemu_ram_foreach_block(cleanup_area, mis)) {
> > +            return -1;
> > +        }
> > +        /*
> > +         * Tell the fault_thread to exit, it's an eventfd that should
> > +         * currently be at 0, we're going to inc it to 1
> > +         */
> > +        tmp64 = 1;
> > +        if (write(mis->userfault_quit_fd, &tmp64, 8) == 8) {
> > +            trace_postcopy_ram_incoming_cleanup_join();
> > +            qemu_thread_join(&mis->fault_thread);
> > +        } else {
> > +            /* Not much we can do here, but may as well report it */
> > +            perror("incing userfault_quit_fd");
> > +        }
> > +        trace_postcopy_ram_incoming_cleanup_closeuf();
> > +        close(mis->userfault_fd);
> > +        close(mis->userfault_quit_fd);
> > +        mis->have_fault_thread = false;
> >      }
> >  
> > +    postcopy_state_set(mis, POSTCOPY_INCOMING_END);
> > +    migrate_send_rp_shut(mis, qemu_file_get_error(mis->file) != 0);
> > +
> >      if (mis->postcopy_tmp_page) {
> >          munmap(mis->postcopy_tmp_page, getpagesize());
> >          mis->postcopy_tmp_page = NULL;
> >      }
> > +    trace_postcopy_ram_incoming_cleanup_exit();
> >      return 0;
> >  }
> >  
> > @@ -531,36 +558,206 @@ static int ram_block_enable_notify(const char 
> > *block_name, void *host_addr,
> >  }
> >  
> >  /*
> > + * Tell the kernel that we've now got some memory it previously asked for.
> > + */
> > +static int ack_userfault(MigrationIncomingState *mis, void *start, size_t 
> > len)
> > +{
> > +    struct uffdio_range range_struct;
> > +
> > +    range_struct.start = (uint64_t)(uintptr_t)start;
> > +    range_struct.len = (uint64_t)len;
> > +
> > +    errno = 0;
> > +    if (ioctl(mis->userfault_fd, UFFDIO_WAKE, &range_struct)) {
> > +        int e = errno;
> > +
> > +        if (e == ENOENT) {
> > +            /* Kernel said it wasn't waiting - one case where this can
> > +             * happen is where two threads triggered the userfault
> > +             * and we receive the page and ack it just after we received
> > +             * the 2nd request and that ends up deciding it should ack it
> > +             * We could optimise it out, but it's rare.
> > +             */
> > +            /*fprintf(stderr, "ack_userfault: %p/%zx ENOENT\n", start, 
> > len); */
> > +            return 0;
> > +        }
> > +        error_report("postcopy_ram: Failed to notify kernel for %p/%zx 
> > (%d)",
> > +                     start, len, e);
> > +        return -e;
> > +    }
> > +
> > +    return 0;
> > +}
> > +
> > +/*
> >   * Handle faults detected by the USERFAULT markings
> >   */
> >  static void *postcopy_ram_fault_thread(void *opaque)
> >  {
> >      MigrationIncomingState *mis = (MigrationIncomingState *)opaque;
> > -
> > -    fprintf(stderr, "postcopy_ram_fault_thread\n");
> > -    /* TODO: In later patch */
> > +    uint64_t hostaddr; /* The kernel always gives us 64 bit, not a pointer 
> > */
> > +    int ret;
> > +    size_t hostpagesize = getpagesize();
> > +    RAMBlock *rb = NULL;
> > +    RAMBlock *last_rb = NULL; /* last RAMBlock we sent part of */
> > +    uint8_t *local_tmp_page;
> > +
> > +    trace_postcopy_ram_fault_thread_entry();
> >      qemu_sem_post(&mis->fault_thread_sem);
> > -    while (1) {
> > -        /* TODO: In later patch */
> > +
> > +    local_tmp_page = mmap(NULL, getpagesize(),
> > +                          PROT_READ | PROT_WRITE, MAP_PRIVATE | 
> > MAP_ANONYMOUS,
> > +                          -1, 0);
> > +    if (!local_tmp_page) {
> > +        perror("mapping local tmp page");
> > +        return NULL;
> >      }
> > +    if (madvise(local_tmp_page, getpagesize(), MADV_DONTFORK)) {
> > +        munmap(local_tmp_page, getpagesize());
> > +        perror("postcpy local page DONTFORK");
> > +        return NULL;
> > +    }
> > +
> > +    while (true) {
> > +        PostcopyPMIState old_state, tmp_state;
> > +        ram_addr_t rb_offset;
> > +        ram_addr_t in_raspace;
> > +        unsigned long bitmap_index;
> > +        struct pollfd pfd[2];
> > +
> > +        /*
> > +         * We're mainly waiting for the kernel to give us a faulting HVA,
> > +         * however we can be told to quit via userfault_quit_fd which is
> > +         * an eventfd
> > +         */
> > +        pfd[0].fd = mis->userfault_fd;
> > +        pfd[0].events = POLLIN;
> > +        pfd[0].revents = 0;
> > +        pfd[1].fd = mis->userfault_quit_fd;
> > +        pfd[1].events = POLLIN; /* Waiting for eventfd to go positive */
> > +        pfd[1].revents = 0;
> > +
> > +        if (poll(pfd, 2, -1 /* Wait forever */) == -1) {
> > +            perror("userfault poll");
> > +            break;
> > +        }
> > +
> > +        if (pfd[1].revents) {
> > +            trace_postcopy_ram_fault_thread_quit();
> > +            break;
> > +        }
> > +
> > +        ret = read(mis->userfault_fd, &hostaddr, sizeof(hostaddr));
> > +        if (ret != sizeof(hostaddr)) {
> > +            if (ret < 0) {
> > +                perror("Failed to read full userfault hostaddr");
> > +                break;
> > +            } else {
> > +                error_report("%s: Read %d bytes from userfaultfd expected 
> > %zd",
> > +                             __func__, ret, sizeof(hostaddr));
> > +                break; /* Lost alignment, don't know what we'd read next */
> > +            }
> > +        }
> > +
> > +        rb = qemu_ram_block_from_host((void *)(uintptr_t)hostaddr, true,
> > +                                      &in_raspace, &rb_offset, 
> > &bitmap_index);
> > +        if (!rb) {
> > +            error_report("postcopy_ram_fault_thread: Fault outside guest: 
> > %"
> > +                         PRIx64, hostaddr);
> > +            break;
> > +        }
> >  
> > +        trace_postcopy_ram_fault_thread_request(hostaddr, bitmap_index,
> > +                                                qemu_ram_get_idstr(rb),
> > +                                                rb_offset);
> > +
> > +        tmp_state = postcopy_pmi_get_state(mis, bitmap_index);
> > +        do {
> > +            old_state = tmp_state;
> > +
> > +            switch (old_state) {
> > +            case POSTCOPY_PMI_REQUESTED:
> > +                /* Do nothing - it's already requested */
> > +                break;
> > +
> > +            case POSTCOPY_PMI_RECEIVED:
> > +                /* Already arrived - no state change, just kick the kernel 
> > */
> > +                trace_postcopy_ram_fault_thread_notify_pre(hostaddr);
> > +                if (ack_userfault(mis,
> > +                                  (void *)((uintptr_t)hostaddr
> > +                                           & ~(hostpagesize - 1)),
> > +                                  hostpagesize)) {
> > +                    assert(0);
> > +                }
> > +                break;
> > +
> > +            case POSTCOPY_PMI_MISSING:
> > +                tmp_state = postcopy_pmi_change_state(mis, bitmap_index,
> > +                                           old_state, 
> > POSTCOPY_PMI_REQUESTED);
> > +                if (tmp_state == POSTCOPY_PMI_MISSING) {
> > +                    /*
> > +                     * Send the request to the source - we want to request 
> > one
> > +                     * of our host page sizes (which is >= TPS)
> > +                     */
> > +                    if (rb != last_rb) {
> > +                        last_rb = rb;
> > +                        migrate_send_rp_req_pages(mis, 
> > qemu_ram_get_idstr(rb),
> > +                                                 rb_offset, hostpagesize);
> > +                    } else {
> > +                        /* Save some space */
> > +                        migrate_send_rp_req_pages(mis, NULL,
> > +                                                 rb_offset, hostpagesize);
> > +                    }
> > +                } /* else it just arrived from the source and the kernel 
> > will
> > +                     be kicked during the receive */
> > +                break;
> > +           }
> > +        } while (tmp_state != old_state);
> 
> Again, I think using separate requested/received bits rather than
> treating them as a single state could avoid this clunky loop.

As mentioned; the PMI is simplified out so this all disappears.

> > +    }
> > +    munmap(local_tmp_page, getpagesize());
> > +    trace_postcopy_ram_fault_thread_exit();
> >      return NULL;
> >  }
> >  
> >  int postcopy_ram_enable_notify(MigrationIncomingState *mis)
> >  {
> > -    /* Create the fault handler thread and wait for it to be ready */
> > +    /* Open the fd for the kernel to give us userfaults */
> > +    mis->userfault_fd = syscall(__NR_userfaultfd, O_CLOEXEC);
> 
> I think it would be good to declare your own userfaultfd() wrappers
> around syscall().  That way it will be easier to clean up once libc
> knows about them.

OK, I'll look at that; this might depend if I find the better way of
dealing with a new syscall definition as mentioned in the other patch.

Dave

> > +    if (mis->userfault_fd == -1) {
> > +        perror("Failed to open userfault fd");
> > +        return -1;
> > +    }
> > +
> > +    /*
> > +     * Although the host check already tested the API, we need to
> > +     * do the check again as an ABI handshake on the new fd.
> > +     */
> > +    if (!ufd_version_check(mis->userfault_fd)) {
> > +        return -1;
> > +    }
> > +
> > +    /* Now an eventfd we use to tell the fault-thread to quit */
> > +    mis->userfault_quit_fd = eventfd(0, EFD_CLOEXEC);
> > +    if (mis->userfault_quit_fd == -1) {
> > +        perror("Opening userfault_quit_fd");
> > +        close(mis->userfault_fd);
> > +        return -1;
> > +    }
> > +
> >      qemu_sem_init(&mis->fault_thread_sem, 0);
> >      qemu_thread_create(&mis->fault_thread, "postcopy/fault",
> >                         postcopy_ram_fault_thread, mis, 
> > QEMU_THREAD_JOINABLE);
> >      qemu_sem_wait(&mis->fault_thread_sem);
> >      qemu_sem_destroy(&mis->fault_thread_sem);
> > +    mis->have_fault_thread = true;
> >  
> >      /* Mark so that we get notified of accesses to unwritten areas */
> >      if (qemu_ram_foreach_block(ram_block_enable_notify, mis)) {
> >          return -1;
> >      }
> >  
> > +    trace_postcopy_ram_enable_notify();
> > +
> >      return 0;
> >  }
> >  
> > diff --git a/trace-events b/trace-events
> > index 16a91d9..d955a28 100644
> > --- a/trace-events
> > +++ b/trace-events
> > @@ -1498,6 +1498,18 @@ postcopy_discard_send_finish(const char *ramblock, 
> > int nwords, int ncmds) "%s ma
> >  postcopy_cleanup_area(const char *ramblock, void *host_addr, size_t 
> > offset, size_t length) "%s: %p offset=%zx length=%zx"
> >  postcopy_init_area(const char *ramblock, void *host_addr, size_t offset, 
> > size_t length) "%s: %p offset=%zx length=%zx"
> >  postcopy_place_page(unsigned long offset, void *host_addr, bool all_zero, 
> > int old_state) "offset=%lx host=%p all_zero=%d old_state=%d"
> > +postcopy_ram_enable_notify(void) ""
> > +postcopy_ram_fault_thread_entry(void) ""
> > +postcopy_ram_fault_thread_exit(void) ""
> > +postcopy_ram_fault_thread_quit(void) ""
> > +postcopy_ram_fault_thread_request(uint64_t hostaddr, unsigned long index, 
> > const char *ramblock, size_t offset) "Request for HVA=%" PRIx64 " index=%lx 
> > rb=%s offset=%zx"
> > +postcopy_ram_fault_thread_notify_pre(uint64_t hostaddr) "%" PRIx64
> > +postcopy_ram_fault_thread_notify_zero(void *hostaddr) "%p"
> > +postcopy_ram_fault_thread_notify_zero_ack(void *hostaddr, unsigned long 
> > bitmap_index) "%p %lx"
> > +postcopy_ram_incoming_cleanup_closeuf(void) ""
> > +postcopy_ram_incoming_cleanup_entry(void) ""
> > +postcopy_ram_incoming_cleanup_exit(void) ""
> > +postcopy_ram_incoming_cleanup_join(void) ""
> >  
> >  # kvm-all.c
> >  kvm_ioctl(int type, void *arg) "type 0x%x, arg %p"
> 
> -- 
> David Gibson                  | I'll have my music baroque, and my code
> david AT gibson.dropbear.id.au        | minimalist, thank you.  NOT _the_ 
> _other_
>                               | _way_ _around_!
> http://www.ozlabs.org/~dgibson


--
Dr. David Alan Gilbert / address@hidden / Manchester, UK