Messages in this thread | | | From | Linus Torvalds <> | Date | Fri, 19 Aug 2011 08:14:02 -0700 | Subject | Re: [build breakage] Re: [GIT PULL] Btrfs pull round two |
| |
On Fri, Aug 19, 2011 at 6:24 AM, Josh Boyer <jwboyer@gmail.com> wrote: > > FYI, this was noticed and a patch was posted to linux-btrfs: > > http://thread.gmane.org/gmane.comp.file-systems.btrfs/12673 > > Josef bugged them to send it upstream.
Oh Christ!
Please don't send that upstream.
This:
- min_free /= 2; + do_div(min_free, 2);
is just moronic. Guys, YOU ARE DIVIDING A UNSIGNED VALUE BY TWO. Has anybody ever looked at a real computer? Doesn't anybody know how computer math works any more? Doing that as a (slow) double division on 32-bit is so stupid that it's past even just "wrong". It's way off in la-la-land, sitting in a corner, all hopped up on drugs and painting its nails purple.
Also, the change that is definitely correct:
- min_free /= dev_min; + do_div(min_free, dev_min);
should still be something that people think about. Why is "dev_min" a signed integer? That looks wrong. You wanted an unsigned divide, didn't you (and that is what "do_div()" is designed for).
So guys, please: when you get an error like this, don't just go into that mindless place where you paper things over. Think about what is going on, and WHY you got the error. Big 64-bit divides are bad bad bad. They are so horrendously bad on 32-bit that we don't even support them (which is the reason I refuse to put that __udiv into the 32-bit libraries), but that are often bad on 64-bit too. So you need to spend some time thinking about it when you get that __udivdi3 error.
Linus
| |