Re: [Qemu-devel] [PATCH V5 7/9] migration: calculate vCPU blocktime on dst side

[Alexey]

On Tue, May 16, 2017 at 12:34:16PM +0100, Dr. David Alan Gilbert wrote:
> * Alexey Perevalov (address@hidden) wrote:
> > This patch provides blocktime calculation per vCPU,
> > as a summary and as a overlapped value for all vCPUs.
> > 
> > This approach was suggested by Peter Xu, as an improvements of
> > previous approch where QEMU kept tree with faulted page address and cpus 
> > bitmask
> > in it. Now QEMU is keeping array with faulted page address as value and vCPU
> > as index. It helps to find proper vCPU at UFFD_COPY time. Also it keeps
> > list for blocktime per vCPU (could be traced with page_fault_addr)
> > 
> > Blocktime will not calculated if postcopy_blocktime field of
> > MigrationIncomingState wasn't initialized.
> > 
> > Signed-off-by: Alexey Perevalov <address@hidden>
> 
> I have some multi-threading/ordering worries still.
> 
> The fault thread receives faults over the ufd and calls
> mark_postcopy_blocktime_being.  That's fine.
> 
> The receiving thread receives pages, calls place page, and
> calls mark_postcopy_blocktime_end.  That's also fine.
> 
> However, remember that we send pages from the source without
> them being requested as background transfers; consider:
> 
> 
>     Source           receive-thread          fault-thread
> 
>   1  Send A
>   2                  Receive A
>   3                                            Access A
>   4                                            Report on UFD
>   5                  Place
>   6                                            Read UFD entry
> 
> 
>  Placing and reading UFD race - and up till now that's been fine;
> so we can read off the ufd an address that's already on it's way from
> the source, and which we might just be receiving, or that we might
> have already placed.
> 
> In this code at (6) won't you call mark_postcopy_blocktime_start
> even though it's already been placed at (5) ? Then that blocktime
> will stay set until the end of the run?
Could you clarify, what does it mean "Read UFD entry",
Place - it's postcopy_place_page.


> 
> Perhaps that's not a problem; if mark_postcopy_blocktime_end is called
> for a different address it wont count the blocktime; and when
> mark_postcopy_blocktime_start is called for a different address it'll
> remove the addres that was a problem above - so perhaps that's fine?
mark_postcopy_blocktime_begin doesn't clear state, it only sets the
state.

Looks like I imaging the race nature:
kernel reports about pagefault for page, which are in the middle of the copying
process, so we will never copy it again, and there is a chance it will
be in vcpu_addr forever.


     Source           receive-thread          fault-thread
 
   4                                            Report on UFD
   5                   Place ioctl(UFFD_COPY)
   5.1                 mark_postcopy_blocktime_end
   4.1                                          mark_postcopy_blocktime_begin

I think that possible, but probability is too low, probability is increasing
for small page size such as 4K pages, ioctl(UFFD_COPY) copies memory
like memcpy doing, so time complexity of copying inside ioctl depends on page
size.
I think to add logic to check *_blocktime_begin in case of 
*_blocktime_end for that page before. Just for not keeping addr in the
vcpu_addr forever.

> 
> 
> > ---
> >  migration/postcopy-ram.c | 87 
> > +++++++++++++++++++++++++++++++++++++++++++++++-
> >  migration/trace-events   |  5 ++-
> >  2 files changed, 90 insertions(+), 2 deletions(-)
> > 
> > diff --git a/migration/postcopy-ram.c b/migration/postcopy-ram.c
> > index a1f1705..e2660ae 100644
> > --- a/migration/postcopy-ram.c
> > +++ b/migration/postcopy-ram.c
> > @@ -23,6 +23,7 @@
> >  #include "migration/postcopy-ram.h"
> >  #include "sysemu/sysemu.h"
> >  #include "sysemu/balloon.h"
> > +#include <sys/param.h>
> >  #include "qemu/error-report.h"
> >  #include "trace.h"
> >  
> > @@ -542,6 +543,86 @@ static int ram_block_enable_notify(const char 
> > *block_name, void *host_addr,
> >      return 0;
> >  }
> >  
> > +static int get_mem_fault_cpu_index(uint32_t pid)
> > +{
> > +    CPUState *cpu_iter;
> > +
> > +    CPU_FOREACH(cpu_iter) {
> > +        if (cpu_iter->thread_id == pid) {
> > +            return cpu_iter->cpu_index;
> > +        }
> > +    }
> > +    trace_get_mem_fault_cpu_index(pid);
> > +    return -1;
> > +}
> > +
> > +static void mark_postcopy_blocktime_begin(uint64_t addr, int cpu)
> > +{
> > +    MigrationIncomingState *mis = migration_incoming_get_current();
> > +    PostcopyBlocktimeContext *dc;
> > +    int64_t now_ms;
> > +    if (!mis->blocktime_ctx || cpu < 0) {
> > +        return;
> > +    }
> 
> You might consider:
> 
>  PostcopyBlocktimeContext *dc = mis->blocktime_ctx;
>  int64_t now_ms;
>  if (!dc || cpu < 0) {
>      return;
>  }
> 
> it gets rid of the two reads of mis->blocktime_ctx
> (You do something similar in a few places)
> 
> > +    now_ms = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
> > +    dc = mis->blocktime_ctx;
> > +    if (dc->vcpu_addr[cpu] == 0) {
> > +        atomic_inc(&dc->smp_cpus_down);
> > +    }
> > +
> > +    atomic_xchg__nocheck(&dc->vcpu_addr[cpu], addr);
> > +    atomic_xchg__nocheck(&dc->last_begin, now_ms);
> > +    atomic_xchg__nocheck(&dc->page_fault_vcpu_time[cpu], now_ms);
> > +
> > +    trace_mark_postcopy_blocktime_begin(addr, dc, 
> > dc->page_fault_vcpu_time[cpu],
> > +            cpu);
> > +}
> > +
> > +static void mark_postcopy_blocktime_end(uint64_t addr)
> > +{
> > +    MigrationIncomingState *mis = migration_incoming_get_current();
> > +    PostcopyBlocktimeContext *dc;
> > +    int i, affected_cpu = 0;
> > +    int64_t now_ms;
> > +    bool vcpu_total_blocktime = false;
> > +
> > +    if (!mis->blocktime_ctx) {
> > +        return;
> > +    }
> > +    dc = mis->blocktime_ctx;
> > +    now_ms = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
> > +
> > +    /* lookup cpu, to clear it,
> > +     * that algorithm looks straighforward, but it's not
> > +     * optimal, more optimal algorithm is keeping tree or hash
> > +     * where key is address value is a list of  */
> > +    for (i = 0; i < smp_cpus; i++) {
> > +        uint64_t vcpu_blocktime = 0;
> > +        if (atomic_fetch_add(&dc->vcpu_addr[i], 0) != addr) {
> > +            continue;
> > +        }
> > +        atomic_xchg__nocheck(&dc->vcpu_addr[i], 0);
> > +        vcpu_blocktime = now_ms -
> > +            atomic_fetch_add(&dc->page_fault_vcpu_time[i], 0);
> > +        affected_cpu += 1;
> > +        /* we need to know is that mark_postcopy_end was due to
> > +         * faulted page, another possible case it's prefetched
> > +         * page and in that case we shouldn't be here */
> > +        if (!vcpu_total_blocktime &&
> > +            atomic_fetch_add(&dc->smp_cpus_down, 0) == smp_cpus) {
> > +            vcpu_total_blocktime = true;
> > +        }
> > +        /* continue cycle, due to one page could affect several vCPUs */
> > +        dc->vcpu_blocktime[i] += vcpu_blocktime;
> > +    }
> > +
> > +    atomic_sub(&dc->smp_cpus_down, affected_cpu);
> > +    if (vcpu_total_blocktime) {
> > +        dc->total_blocktime += now_ms - atomic_fetch_add(&dc->last_begin, 
> > 0);
> 
> This total_blocktime calculation is a little odd; the 'last_begin' is
> not necessarily related to the same CPU or same block.
last_begin should not be related to the same vCPU, vCPU doesn't
matter in this case, due to last_begin is a time when
mark_postcopy_blocktime_begin was called (last pagefault),
so if we 100% sure here all vCPU is blocked, the time interval since it
was blocked starts at last_begin time, even that last_begin was on
another vCPU.

> 
> Dave
> 
> > +    }
> > +    trace_mark_postcopy_blocktime_end(addr, dc, dc->total_blocktime);
> > +}
> > +
> >  /*
> >   * Handle faults detected by the USERFAULT markings
> >   */
> > @@ -619,8 +700,11 @@ static void *postcopy_ram_fault_thread(void *opaque)
> >          rb_offset &= ~(qemu_ram_pagesize(rb) - 1);
> >          trace_postcopy_ram_fault_thread_request(msg.arg.pagefault.address,
> >                                                  qemu_ram_get_idstr(rb),
> > -                                                rb_offset);
> > +                                                rb_offset,
> > +                                                
> > msg.arg.pagefault.feat.ptid);
> >  
> > +        
> > mark_postcopy_blocktime_begin((uintptr_t)(msg.arg.pagefault.address),
> > +                         
> > get_mem_fault_cpu_index(msg.arg.pagefault.feat.ptid));
> >          /*
> >           * Send the request to the source - we want to request one
> >           * of our host page sizes (which is >= TPS)
> > @@ -715,6 +799,7 @@ int postcopy_place_page(MigrationIncomingState *mis, 
> > void *host, void *from,
> >  
> >          return -e;
> >      }
> > +    mark_postcopy_blocktime_end((uint64_t)(uintptr_t)host);
> >  
> >      trace_postcopy_place_page(host);
> >      return 0;
> > diff --git a/migration/trace-events b/migration/trace-events
> > index b8f01a2..9424e3e 100644
> > --- a/migration/trace-events
> > +++ b/migration/trace-events
> > @@ -110,6 +110,8 @@ process_incoming_migration_co_end(int ret, int ps) 
> > "ret=%d postcopy-state=%d"
> >  process_incoming_migration_co_postcopy_end_main(void) ""
> >  migration_set_incoming_channel(void *ioc, const char *ioctype) "ioc=%p 
> > ioctype=%s"
> >  migration_set_outgoing_channel(void *ioc, const char *ioctype, const char 
> > *hostname)  "ioc=%p ioctype=%s hostname=%s"
> > +mark_postcopy_blocktime_begin(uint64_t addr, void *dd, int64_t time, int 
> > cpu) "addr 0x%" PRIx64 " dd %p time %" PRId64 " cpu %d"
> > +mark_postcopy_blocktime_end(uint64_t addr, void *dd, int64_t time) "addr 
> > 0x%" PRIx64 " dd %p time %" PRId64
> >  
> >  # migration/rdma.c
> >  qemu_rdma_accept_incoming_migration(void) ""
> > @@ -186,7 +188,7 @@ 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, const char *ramblock, 
> > size_t offset) "Request for HVA=%" PRIx64 " rb=%s offset=%zx"
> > +postcopy_ram_fault_thread_request(uint64_t hostaddr, const char *ramblock, 
> > size_t offset, uint32_t pid) "Request for HVA=%" PRIx64 " rb=%s offset=%zx 
> > %u"
> >  postcopy_ram_incoming_cleanup_closeuf(void) ""
> >  postcopy_ram_incoming_cleanup_entry(void) ""
> >  postcopy_ram_incoming_cleanup_exit(void) ""
> > @@ -195,6 +197,7 @@ save_xbzrle_page_skipping(void) ""
> >  save_xbzrle_page_overflow(void) ""
> >  ram_save_iterate_big_wait(uint64_t milliconds, int iterations) "big wait: 
> > %" PRIu64 " milliseconds, %d iterations"
> >  ram_load_complete(int ret, uint64_t seq_iter) "exit_code %d seq iteration 
> > %" PRIu64
> > +get_mem_fault_cpu_index(uint32_t pid) "pid %u is not vCPU"
> >  
> >  # migration/exec.c
> >  migration_exec_outgoing(const char *cmd) "cmd=%s"
> > -- 
> > 1.9.1
> > 
> --
> Dr. David Alan Gilbert / address@hidden / Manchester, UK
> 

-- 

BR
Alexey