Messages in this thread | | | Date | Tue, 13 Dec 2016 07:05:35 -0800 | From | "Paul E. McKenney" <> | Subject | Re: [PATCH] rcu: shift by 1UL rather than 1 to fix sign extension error |
| |
On Tue, Dec 13, 2016 at 08:25:42PM +0800, Boqun Feng wrote: > On Tue, Dec 13, 2016 at 07:21:48PM +0800, Boqun Feng wrote: > > On Tue, Dec 13, 2016 at 10:56:46AM +0000, Colin King wrote: > > > From: Colin Ian King <colin.king@canonical.com> > > > > > > mask and bit are unsigned longs, so if bit is 31 we end up sign > > > extending the 1 and mask ends up as 0xffffffff80000000. Fix this > > > by explicitly adding integer suffix UL ensure 1 is a unsigned long > > > rather than an signed int. > > > > > > > Right, you are, and the tool is ;-) > > > > If @bit is greater than 32, we even got an undefined behavior in C ;-( > > This is my careless mistake, thank you for finding it out and fix it! > > > > > Issue found with static analysis with CoverityScan, CID 1388564 > > > > > > Fixes: 8965c3ce4718754db ("rcu: Use leaf_node_for_each_mask_possible_cpu() in force_qs_rnp()") > > > Signed-off-by: Colin Ian King <colin.king@canonical.com> > > > > I think Paul only queued that for running tests and I have almost > > finished a v2. I will fold your fix in my patch and add your SoB along > > with mine, does that work for you? > > > > TBH, this situation is kinda new to me, so if anyone has any suggestion, > > please let me know ;-) > > > > Regards, > > Boqun > > > > > --- > > > kernel/rcu/tree.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c > > > index 10162ac..6ecedd8 100644 > > > --- a/kernel/rcu/tree.c > > > +++ b/kernel/rcu/tree.c > > > @@ -3051,7 +3051,7 @@ static void force_qs_rnp(struct rcu_state *rsp, > > > > > > leaf_node_for_each_mask_possible_cpu(rnp, rnp->qsmask, bit, cpu) > > > if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj)) > > > - mask |= 1 << bit; > > > + mask |= 1UL << bit; > > Hmm.. Seems using BIT() here is a good idea, and maybe rename bit as > grp_idx or something.
Well, "bit" could be a bit, or it could be the number of the bit. Given that we have "mask", and given that we are shifting by it, it has to be the number of the bit.
> Naming, naming, naming..
Often I think that the biggest problem with naming is putting too much time into worrying about it. ;-)
But if you want to change the name from "bit" to "bitno" or "bitnum" as part of your updated patchset, I won't object. The reason for preferring either to "grp_idx" is that "bitno" goes well with "mask". But as a general rule, I must follow the usual practice of not favoring renaming patches.
Thanx, Paul
> Regards, > Boqun > > > > > > > if (mask != 0) { > > > /* Idle/offline CPUs, report (releases rnp->lock. */ > > > -- > > > 2.10.2 > > > > >
| |