Am 31.05.2023 um 13:02 hat Kevin Wolf geschrieben:
> blk_insert_bs() requires that callers hold the AioContext lock for the
> node that should be inserted. Take it.
>
> Signed-off-by: Kevin Wolf
> ---
> hw/core/qdev-properties-system.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/hw/core/qdev-properties-system.c
> b/hw/core/qdev-properties-system.c
> index d42493f630..7f6b14276a 100644
> --- a/hw/core/qdev-properties-system.c
> +++ b/hw/core/qdev-properties-system.c
> @@ -147,7 +147,10 @@ static void set_drive_helper(Object *obj, Visitor *v,
> const char *name,
> blk = blk_new(ctx, 0, BLK_PERM_ALL);
> blk_created = true;
>
> +aio_context_acquire(ctx);
> ret = blk_insert_bs(blk, bs, errp);
> +aio_context_release(ctx);
This one actually isn't completely right. ctx is only the AioContext of
bs if the device supports iothreads, but the main context otherwise.
It's the right thing for blk_new(), but for blk_insert_bs() we always
need to hold the lock for the AioContext of bs.
Kevin