On 08/07/2015 21:13, Kevin Wolf wrote: > Am 07.07.2015 um 15:09 hat Juan Quintela geschrieben: >> From: Li Zhijian <lizhij...@cn.fujitsu.com> >> >> Signed-off-by: Li Zhijian <lizhij...@cn.fujitsu.com> >> Signed-off-by: Wen Congyang <we...@cn.fujitsu.com> >> Signed-off-by: Juan Quintela <quint...@redhat.com> >> --- >> migration/ram.c | 23 +++++++++++++++++------ >> 1 file changed, 17 insertions(+), 6 deletions(-) > > In current master, HMP 'savevm' is broken (looks like a deadlock in RCU > code, it just hangs indefinitely). git bisect points to this patch.
This looks like synchronize_rcu() is being called within rcu_read_lock()/rcu_read_unlock(). The easiest fix is to somehow use call_rcu, but I haven't looked at the code very well. I found another embarrassing bug in the RCU code, but it's been there forever and can wait for after -rc0 (and it wasn't really a problem until BQL-less MMIO was merged a couple days ago). Paolo > The stack trace looks like this: > > (gdb) thread apply all bt > > Thread 3 (Thread 0x7f06febfe700 (LWP 5717)): > #0 0x00007f070e749f7d in __lll_lock_wait () from /lib64/libpthread.so.0 > #1 0x00007f070e745d32 in _L_lock_791 () from /lib64/libpthread.so.0 > #2 0x00007f070e745c38 in pthread_mutex_lock () from /lib64/libpthread.so.0 > #3 0x00007f070fed8bc9 in qemu_mutex_lock (mutex=mutex@entry=0x7f07107e6700 > <rcu_gp_lock>) at util/qemu-thread-posix.c:73 > #4 0x00007f070fee7631 in synchronize_rcu () at util/rcu.c:129 > #5 0x00007f070fee77d9 in call_rcu_thread (opaque=<optimized out>) at > util/rcu.c:240 > #6 0x00007f070e743df5 in start_thread () from /lib64/libpthread.so.0 > #7 0x00007f07066ab1ad in clone () from /lib64/libc.so.6 > > Thread 2 (Thread 0x7f06f940f700 (LWP 5719)): > #0 0x00007f070e749f7d in __lll_lock_wait () from /lib64/libpthread.so.0 > #1 0x00007f070e74c4ec in _L_cond_lock_792 () from /lib64/libpthread.so.0 > #2 0x00007f070e74c3c8 in __pthread_mutex_cond_lock () from > /lib64/libpthread.so.0 > #3 0x00007f070e747795 in pthread_cond_wait@@GLIBC_2.3.2 () from > /lib64/libpthread.so.0 > #4 0x00007f070fed8ca9 in qemu_cond_wait (cond=<optimized out>, > mutex=mutex@entry=0x7f07103b0400 <qemu_global_mutex>) at > util/qemu-thread-posix.c:132 > #5 0x00007f070fc4daab in qemu_tcg_cpu_thread_fn (arg=<optimized out>) at > /home/kwolf/source/qemu/cpus.c:1050 > #6 0x00007f070e743df5 in start_thread () from /lib64/libpthread.so.0 > #7 0x00007f07066ab1ad in clone () from /lib64/libc.so.6 > > Thread 1 (Thread 0x7f070fb20bc0 (LWP 5716)): > #0 0x00007f07066a5949 in syscall () from /lib64/libc.so.6 > #1 0x00007f070fed8fa2 in futex_wait (val=4294967295, ev=0x7f07107e66c0 > <rcu_gp_event>) at util/qemu-thread-posix.c:301 > #2 qemu_event_wait (ev=ev@entry=0x7f07107e66c0 <rcu_gp_event>) at > util/qemu-thread-posix.c:399 > #3 0x00007f070fee7713 in wait_for_readers () at util/rcu.c:120 > #4 synchronize_rcu () at util/rcu.c:149 > #5 0x00007f070fc6e0c2 in migration_end () at > /home/kwolf/source/qemu/migration/ram.c:1033 > #6 0x00007f070fc6ef23 in ram_save_complete (f=0x7f07122f9aa0, > opaque=<optimized out>) at /home/kwolf/source/qemu/migration/ram.c:1241 > #7 0x00007f070fc71d75 in qemu_savevm_state_complete > (f=f@entry=0x7f07122f9aa0) at /home/kwolf/source/qemu/migration/savevm.c:836 > #8 0x00007f070fc7298e in qemu_savevm_state (errp=0x7ffe2a081ff8, > f=0x7f07122f9aa0) at /home/kwolf/source/qemu/migration/savevm.c:945 > #9 hmp_savevm (mon=0x7f071233b500, qdict=<optimized out>) at > /home/kwolf/source/qemu/migration/savevm.c:1353 > #10 0x00007f070fc552d0 in handle_hmp_command (mon=mon@entry=0x7f071233b500, > cmdline=0x7f0712350197 "foo") at /home/kwolf/source/qemu/monitor.c:4058 > #11 0x00007f070fc56467 in monitor_command_cb (opaque=0x7f071233b500, > cmdline=<optimized out>, readline_opaque=<optimized out>) > at /home/kwolf/source/qemu/monitor.c:5081 > #12 0x00007f070fee6dbf in readline_handle_byte (rs=0x7f0712350190, > ch=<optimized out>) at util/readline.c:391 > #13 0x00007f070fc55387 in monitor_read (opaque=<optimized out>, > buf=<optimized out>, size=<optimized out>) at > /home/kwolf/source/qemu/monitor.c:5064 > #14 0x00007f070fd17b21 in qemu_chr_be_write (len=<optimized out>, > buf=0x7ffe2a082640 "\n\331\367\325\001\200\377\377\200&\b*\376\177", > s=0x7f0712304670) > at qemu-char.c:306 > #15 fd_chr_read (chan=<optimized out>, cond=<optimized out>, > opaque=0x7f0712304670) at qemu-char.c:1012 > #16 0x00007f070e04c9ba in g_main_context_dispatch () from > /lib64/libglib-2.0.so.0 > #17 0x00007f070fe61678 in glib_pollfds_poll () at main-loop.c:199 > #18 os_host_main_loop_wait (timeout=<optimized out>) at main-loop.c:244 > #19 main_loop_wait (nonblocking=<optimized out>) at main-loop.c:493 > #20 0x00007f070fc24e9e in main_loop () at vl.c:1901 > #21 main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) > at vl.c:4649 > > Kevin > >> diff --git a/migration/ram.c b/migration/ram.c >> index 644f52a..9c0bcfe 100644 >> --- a/migration/ram.c >> +++ b/migration/ram.c >> @@ -494,6 +494,7 @@ static int save_xbzrle_page(QEMUFile *f, uint8_t >> **current_data, >> return 1; >> } >> >> +/* Called with rcu_read_lock() to protect migration_bitmap */ >> static inline >> ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr, >> ram_addr_t start) >> @@ -502,26 +503,31 @@ ram_addr_t >> migration_bitmap_find_and_reset_dirty(MemoryRegion *mr, >> unsigned long nr = base + (start >> TARGET_PAGE_BITS); >> uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr)); >> unsigned long size = base + (mr_size >> TARGET_PAGE_BITS); >> + unsigned long *bitmap; >> >> unsigned long next; >> >> + bitmap = atomic_rcu_read(&migration_bitmap); >> if (ram_bulk_stage && nr > base) { >> next = nr + 1; >> } else { >> - next = find_next_bit(migration_bitmap, size, nr); >> + next = find_next_bit(bitmap, size, nr); >> } >> >> if (next < size) { >> - clear_bit(next, migration_bitmap); >> + clear_bit(next, bitmap); >> migration_dirty_pages--; >> } >> return (next - base) << TARGET_PAGE_BITS; >> } >> >> +/* Called with rcu_read_lock() to protect migration_bitmap */ >> static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length) >> { >> + unsigned long *bitmap; >> + bitmap = atomic_rcu_read(&migration_bitmap); >> migration_dirty_pages += >> - cpu_physical_memory_sync_dirty_bitmap(migration_bitmap, start, >> length); >> + cpu_physical_memory_sync_dirty_bitmap(bitmap, start, length); >> } >> >> >> @@ -1017,10 +1023,15 @@ void free_xbzrle_decoded_buf(void) >> >> static void migration_end(void) >> { >> - if (migration_bitmap) { >> + /* caller have hold iothread lock or is in a bh, so there is >> + * no writing race against this migration_bitmap >> + */ >> + unsigned long *bitmap = migration_bitmap; >> + atomic_rcu_set(&migration_bitmap, NULL); >> + if (bitmap) { >> memory_global_dirty_log_stop(); >> - g_free(migration_bitmap); >> - migration_bitmap = NULL; >> + synchronize_rcu(); >> + g_free(bitmap); >> } >> >> XBZRLE_cache_lock(); >> -- >> 2.4.3 >> >> > >