On 11/1/21 11:02 AM, Linus Torvalds wrote: > On Sun, Oct 31, 2021 at 12:41 PM Jens Axboe <axboe@xxxxxxxxx> wrote: >> >> On top of the core block branch, this topic branch cleans up the bdev >> size handling. > > So on the whole this seems to be a good cleanup, but some of it worries me. > > For example, it seems to have lost the cast to "loff_t" when > generating the byte size from a "sector_t". > > Ok, so these days those are both 64-bit, and it doesn't actually > matter (the time when we had a 32-bit sector_t as an option are long > gone), but I think that bdev_nr_bytes() helper really ends up being > subtler than it looks. It very much depends on 'sector_t' and 'loff_t' > being the same size (although sector_t is an u64, loff_t ends up being > the signed version). > > I've pulled this, but I do think it might have been better with the > type conversion being explicit. One of the reasons we had "sector_t" > originally was that it ended up being configuration-dependent, and > could be 32-bit. Those times may be gone, but it's still conceptually > a very different type from "loff_t". Yes, probably safer just to make bdev_nr_bytes() return sector_t as well, even if loff_t isn't strictly wrong. Christoph, want to do a followup? -- Jens Axboe
