----- Original Message ----- > From: "Fam Zheng" <f...@redhat.com> > To: "Paolo Bonzini" <pbonz...@redhat.com> > Cc: "QEMU Developers" <qemu-devel@nongnu.org> > Sent: Thursday, January 25, 2018 4:05:27 PM > Subject: Re: [Qemu-devel] [PATCH v2 0/4] coroutine-lock: polymorphic CoQueue > > On Tue, Jan 16, 2018 at 10:23 PM, Paolo Bonzini <pbonz...@redhat.com> wrote: > > There are cases in which a queued coroutine must be restarted from > > non-coroutine context (with qemu_co_enter_next). In this cases, > > qemu_co_enter_next also needs to be thread-safe, but it cannot use a > > CoMutex and so cannot qemu_co_queue_wait. This happens in curl (which > > right now is rolling its own list of Coroutines) and will happen in > > Fam's NVMe driver as well. > > > > This series extracts the idea of a polymorphic lockable object > > from my "scoped lock guard" proposal, and applies it to CoQueue. > > The implementation of QemuLockable is similar to C11 _Generic, but > > redone using the preprocessor and GCC builtins for compatibility. > > > > In general, while a bit on the esoteric side, the functionality used > > to emulate _Generic is fairly old in GCC, and the builtins are already > > used by include/qemu/atomic.h; the series was tested with Fedora 27 (boot > > Damn Small Linux via http) and CentOS 6 (compiled only). > > I'm seeing this crash with the series: > > (gdb) bt > #0 0x00007ff76204d66b in raise () at /lib64/libc.so.6 > #1 0x00007ff76204f381 in abort () at /lib64/libc.so.6 > #2 0x00007ff7620458fa in __assert_fail_base () at /lib64/libc.so.6 > #3 0x00007ff762045972 in () at /lib64/libc.so.6 > #4 0x000055eaab249c68 in qemu_co_mutex_unlock (mutex=0x7ff750bf7b40) > at /stor/work/qemu/util/qemu-coroutine-lock.c:320 > #5 0x000055eaab249da3 in qemu_lockable_unlock (x=0x7ff750bf7b40) at > /stor/work/qemu/include/qemu/lockable.h:72 > #6 0x000055eaab249da3 in qemu_co_queue_wait_impl > (queue=0x55eaaef41a08, lock=lock@entry=0x7ff750bf7b40) at > /stor/work/qemu/util/qemu-coroutine-lock.c:49 > #7 0x000055eaab19f2b9 in handle_dependencies > (bs=bs@entry=0x55eaad9c6620, guest_offset=guest_offset@entry=1597440, > cur_bytes=cur_bytes@entry=0x7ff750bf7ba0, m=m@entry=0x7ff7 > 50bf7c58) at /stor/work/qemu/block/qcow2-cluster.c:1067 > #8 0x000055eaab1a1b85 in qcow2_alloc_cluster_offset > (bs=bs@entry=0x55eaad9c6620, offset=offset@entry=1597440, > bytes=bytes@entry=0x7ff750bf7c4c, host_offset=host_offset@entry > =0x7ff750bf7c50, m=m@entry=0x7ff750bf7c58) at > /stor/work/qemu/block/qcow2-cluster.c:1497 > #9 0x000055eaab19411e in qcow2_co_pwritev (bs=0x55eaad9c6620, > offset=1597440, bytes=8192, qiov=0x55eaaedb4880, flags=<optimized > out>) at /stor/work/qemu/block/qcow2.c:1896 > #10 0x000055eaab1c2962 in bdrv_driver_pwritev > (bs=bs@entry=0x55eaad9c6620, offset=offset@entry=1597440, > bytes=bytes@entry=8192, qiov=qiov@entry=0x55eaaedb4880, flags=flags@en > try=0) at /stor/work/qemu/block/io.c:976 > #11 0x000055eaab1c3985 in bdrv_aligned_pwritev > (child=child@entry=0x55eaad92bd00, req=req@entry=0x7ff750bf7e70, > offset=offset@entry=1597440, bytes=bytes@entry=8192, align=ali > gn@entry=1, qiov=qiov@entry=0x55eaaedb4880, flags=0) at > /stor/work/qemu/block/io.c:1534 > #12 0x000055eaab1c4ca5 in bdrv_co_pwritev (child=0x55eaad92bd00, > offset=offset@entry=1597440, bytes=bytes@entry=8192, > qiov=qiov@entry=0x55eaaedb4880, flags=flags@entry=0) > at /stor/work/qemu/block/io.c:1785 > #13 0x000055eaab1b4f06 in blk_co_pwritev (blk=0x55eaad9c63c0, > offset=1597440, bytes=8192, qiov=0x55eaaedb4880, flags=0) at > /stor/work/qemu/block/block-backend.c:1135 > #14 0x000055eaab1b4fff in blk_aio_write_entry (opaque=0x55eaaefc5eb0) > at /stor/work/qemu/block/block-backend.c:1326 > #15 0x000055eaab24a77a in coroutine_trampoline (i0=<optimized out>, > i1=<optimized out>) at /stor/work/qemu/util/coroutine-ucontext.c:79 > #16 0x00007ff762066bc0 in __start_context () at /lib64/libc.so.6 > #17 0x00007ffdf69102d0 in () > #18 0x0000000000000000 in () > > It's late today so I'll take a closer look tomorrow.
Ouch. /me bangs the head against the wall and runs writing testcases. Sorry. Paolo
