On 5/19/25 10:54 PM, Ming Lei wrote:
> @@ -2014,6 +2064,14 @@ static int ublk_commit_and_fetch(const struct ublk_queue *ubq,
> return -EINVAL;
> }
>
> + if (ublk_support_auto_buf_reg(ubq)) {
> + if (io->flags & UBLK_IO_FLAG_AUTO_BUF_REG) {
> + WARN_ON_ONCE(io_buffer_unregister_bvec(cmd, 0,
> + issue_flags));
> + io->flags &= ~UBLK_IO_FLAG_AUTO_BUF_REG;
> + }
> + }
> +
Debug or WARN_ON_ONCE() statements with side effects is generally not a
good idea, imho. Would be cleaner to do:
ret = io_buffer_unregister_bvec(cmd, 0, issue_flags);
WARN_ON_ONCE(ret);
and ditto for the next patch that then updates it.
But pretty minor, not worth a respin. This series looks ready to me at
this point.
--
Jens Axboe
