From: Denis V. Lunev <[email protected]> Every user of the cpr fd table runs under the BQL: device realize and unrealize, backend creation, RAM block allocation and freeing (the latter also from RCU callbacks, which take the BQL), vfio MSI-X vector setup from vCPU threads, and the cpr state save, load and cpr-exec walk on the main thread. The table has no lock of its own and depends on that, but nothing says so.
Say it at the declarations. The obvious way to break it is a socket chardev preserved across CPR: its connection hooks run in the chardev's GMainContext, which for the monitor is the QMP iothread, and that thread must not take the BQL or out-of-band commands stop working. Such a user needs the table to grow a lock of its own, and has to keep the saved table and the preserved fds in agreement when a connection comes or goes during cpr-exec. Both belong with the series that adds it. Signed-off-by: Denis V. Lunev <[email protected]> Cc: Maciej S. Szmigiero <[email protected]> Cc: Peter Xu <[email protected]> Cc: Fabiano Rosas <[email protected]> Cc: Mark Kanda <[email protected]> Cc: Ben Chaney <[email protected]> Cc: Steve Sistare <[email protected]> --- v2: replace the lock with a comment, as discussed on v1: https://lore.kernel.org/qemu-devel/[email protected]/ include/migration/cpr.h | 1 + 1 file changed, 1 insertion(+) diff --git a/include/migration/cpr.h b/include/migration/cpr.h index 3ad084d5ba..5ac46b6207 100644 --- a/include/migration/cpr.h +++ b/include/migration/cpr.h @@ -28,6 +28,7 @@ typedef struct CprState { extern CprState cpr_state; +/* The cpr fd table, including cpr_walk_fd(), is protected by the BQL. */ bool cpr_save_fd(const char *name, int id, int fd, Error **errp); void cpr_delete_fd(const char *name, int id); int cpr_find_fd(const char *name, int id); base-commit: f29ccc287a6cfdd51d3a4fd17fbe9af3310a0ac5 -- 2.53.0
