On 3/27/23 2:02 PM, Al Viro wrote: > On Mon, Mar 27, 2023 at 12:59:09PM -0600, Jens Axboe wrote: > >>> That's a great idea. Two questions - do we want to make that >>> WARN_ON_ONCE()? And then do we want to include a WARN_ON_ONCE for a >>> non-supported type? Doesn't seem like high risk as they've all been used >>> with ITER_IOVEC until now, though. >> >> Scratch that last one, user_backed should double as that as well. At >> least currently, where ITER_UBUF and ITER_IOVEC are the only two >> iterators that hold user backed memory. > > Quite. As for the WARN_ON_ONCE vs. WARN_ON... No preferences, really. OK, I'll stick with WARN_ON_ONCE then. At least that avoids a ton of dmesg dumping for something buggy, yet still preserving the first trace. -- Jens Axboe
