On Mon, Oct 05, 2020 at 03:12:22PM -0700, Paul E. McKenney wrote:
>On Sat, Oct 03, 2020 at 09:19:25AM -0700, Paul E. McKenney wrote:
>> On Sat, Oct 03, 2020 at 12:28:47PM +0800, Junchang Wang wrote:
>
>[ . . . ]
>
>> But sadly, this change does not fix this bug:
>>
>> $ ./hash_bkt_hazptr --schroedinger --nreaders 1 --duration 1000 --updatewait 0 --nbuckets 8192 --elems/writer 8192
>> Segmentation fault (core dumped)
>>
>> Can't have everything. ;-)
>>
>> It appears that hashtorture manages to fail to call the per-thread
>> hazard-pointers initialization on some paths, so it is no surprise
>> that Junchang's fix doesn't fix this bug.
>
> Thanx, Paul
>
>------------------------------------------------------------------------
>
>commit 2533decebb7fca65d2860174246966f5ed255dd0
>Author: Paul E. McKenney <paulmck@xxxxxxxxxx>
>Date: Mon Oct 5 15:10:03 2020 -0700
>
> datastruct/hash: Add missing hash_register_thread()
>
> The zoo_test() function was missing the hash_register_thread() invocation
> that is required to make hazard pointers work correctly. The lack
> of this invocation results in a segmentation violation. This commit
> therefore adds this call.
>
> Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
>
Hi Paul,
This patch fixes the bug, and now "./hash_bkt_hazptr --schroedinger" works
correctly on my servers. Thanks a lot.
--Junchang
>diff --git a/CodeSamples/datastruct/hash/hashtorture.h b/CodeSamples/datastruct/hash/hashtorture.h
>index 6d30cb8..7d43e11 100644
>--- a/CodeSamples/datastruct/hash/hashtorture.h
>+++ b/CodeSamples/datastruct/hash/hashtorture.h
>@@ -1211,6 +1211,7 @@ void zoo_test(void)
> for (i = 0; i < nupdaters * elperupdater; i++) {
> sprintf(&zoo_names[ZOO_NAMELEN * i], "a%ld", i);
> }
>+ hash_register_thread();
>
> zhep = malloc(sizeof(*zhep));
> BUG_ON(!zhep);
