Re: [Qemu-devel] [PATCH] gdb command: qemu iohandlers

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Fam Zheng (famz@redhat.com) wrote:
> On Wed, 06/24 16:19, Stefan Hajnoczi wrote:
> > On Tue, Jun 23, 2015 at 03:43:53PM +0100, Dr. David Alan Gilbert (git) wrote:
> > > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
> > > 
> > > Add a gdb command to print the current set of IOHandlers and
> > > if one of them is a thread yielding for data print the backtrace.
> > > 
> > > Useful for debugging why an incoming migration has stalled, e.g.
> > > 
> > >   ----
> > >   {fd_read = 0x7fd4c8e40d00 <fd_coroutine_enter>, fd_write = 0x0, opaque =
> > >   0x7fd4b8bfeb00, next = {le_next = 0x7fd4cac81b00, le_prev =
> > >   0x7fd4c93d2bd0 <io_handlers>}, fd = 22, pollfds_idx = 0, deleted =
> > >   false}
> > >   #0  qemu_coroutine_switch (from_=from_@entry=0x7fd4cbb33c00,
> > >   to_=to_@entry=0x7fd4c8b13a90,
> > >       action=action@entry=COROUTINE_YIELD) at coroutine-ucontext.c:177
> > >   #1  0x00007fd4c8e40507 in qemu_coroutine_yield () at
> > >   qemu-coroutine.c:145
> > >   #2  0x00007fd4c8e40e75 in yield_until_fd_readable (fd=22) at
> > >   qemu-coroutine-io.c:90
> > >   #3  0x00007fd4c8df347f in qemu_rdma_block_for_wrid
> > >   (rdma=rdma@entry=0x7fd4b8c7e010,
> > >       wrid_requested=wrid_requested@entry=2000,
> > >   byte_len=byte_len@entry=0x0) at migration/rdma.c:1510
> > >   #4  0x00007fd4c8df388f in qemu_rdma_post_send_control
> > >   (rdma=rdma@entry=0x7fd4b8c7e010, buf=buf@entry=0x0,
> > >       head=head@entry=0x7fd4b8bfed00) at migration/rdma.c:1608
> > >   #5  0x00007fd4c8df4b8e in qemu_rdma_exchange_recv (rdma=0x7fd4b8c7e010,
> > >   head=0x7fd4b8bfed50, expecting=3)
> > >       at migration/rdma.c:1814
> > >   #6  0x00007fd4c8df5089 in qemu_rdma_get_buffer (opaque=0x7fd4cba34950,
> > >   buf=0x7fd4cc24fd20 "TR\022",
> > >       pos=<optimized out>, size=32768) at migration/rdma.c:2611
> > >   #7  0x00007fd4c8df000d in qemu_fill_buffer (f=f@entry=0x7fd4cc24fcf0) at
> > >   migration/qemu-file.c:214
> > >   #8  0x00007fd4c8df08d4 in qemu_peek_byte (f=f@entry=0x7fd4cc24fcf0,
> > >   offset=offset@entry=0)
> > >       at migration/qemu-file.c:447
> > >   #9  0x00007fd4c8c2cad1 in qemu_loadvm_state (f=f@entry=0x7fd4cc24fcf0)
> > >   at /root/qemu-world3/migration/savevm.c:1128
> > >   #10 0x00007fd4c8ded895 in process_incoming_migration_co
> > >   (opaque=0x7fd4cc24fcf0) at migration/migration.c:145
> > >   #11 0x00007fd4c8e4112a in coroutine_trampoline (i0=<optimized out>,
> > >   i1=<optimized out>) at coroutine-ucontext.c:80
> > >   #12 0x00007fd4c453e0f0 in ?? () from /lib64/libc.so.6
> > >   #13 0x00007ffee263e870 in ?? ()
> > >   #14 0x0000000000000000 in ?? ()
> > >   ----
> > >   {fd_read = 0x7fd4c8e3ace0 <sigfd_handler>, fd_write = 0x0, opaque = 0x5,
> > >   next = {le_next = 0x0, le_prev = 0x7fd4cb78a3d8}, fd = 5, pollfds_idx =
> > >   1, deleted = false}
> > > 
> > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
> > > ---
> > >  scripts/qemu-gdb.py | 25 +++++++++++++++++++++++++
> > >  1 file changed, 25 insertions(+)
> > > 
> > > diff --git a/scripts/qemu-gdb.py b/scripts/qemu-gdb.py
> > > index 6c7f4fb..2670625 100644
> > > --- a/scripts/qemu-gdb.py
> > > +++ b/scripts/qemu-gdb.py
> > > @@ -102,6 +102,30 @@ class CoroutineCommand(gdb.Command):
> > >          coroutine_pointer = gdb.parse_and_eval(argv[0]).cast(gdb.lookup_type('CoroutineUContext').pointer())
> > >          bt_jmpbuf(coroutine_pointer['env']['__jmpbuf'])
> > >  
> > > +class IOhandlersCommand(gdb.Command):
> > > +    '''Display a backtrace for io handlers'''
> > > +    def __init__(self):
> > > +        gdb.Command.__init__(self, 'qemu iohandlers', gdb.COMMAND_DATA,
> > > +                             gdb.COMPLETE_NONE)
> > > +
> > > +    def invoke(self, arg, from_tty):
> > > +        cur = gdb.parse_and_eval('io_handlers.lh_first')
> > 
> > Fam is getting rid of io_handlers and replacing it with an AioContext.
> > 
> 
> Yes, the series is
> 
>   [PATCH 0/9] slirp: iohandler: Rebase onto aio

OK, I can rework after that goes in.
I guess the equivalent there is to dump the iohandler_ctx->aio_handlers list?

Dave

> 
> Fam
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK