[Jamvm-general] Hash entry scavenge problem.

[Dore <do...@do...>]

Hi,=20

Yesterday I posted the message below as a bug on sourceforge, but from one
of my collegues I just heard that Nabble is the place to be if the issue is
JamVM. Therefor I also post it here.

We are running jamVM on an platform with a limited amount of mutexes and
due to that we have found minor error in the for-loop used for scavenging
data in the hash-table.

The error will not show on platforms with "unlimited" mutexes and is
therefor probably not critical on most platforms.

In hash.h line 97-98 there is a for-loop:

int n;
for (i=3D0, n=3Dtable.hash_count;n--;i++) {

When using the macro in lock.c it loops through the mon_cache an moves some
of the monitors over to a free_list. But as it only count the number of
actually used entries it never "cleans" the top area of the mon_cache and
therefore the hash_table is resized eventhough there is a number of entries
that could have been reused. This leads to a leak in allocated mutexs and
finally a system lock. We have fixed the problem in our application by
correcting the for-loop to:

for (i=3D0;i < table.hash_size;i++) {


Best regards
Mads Dor=C3=A9
dore at doredevelopment.dk

--=20
View this message in context: http://www.nabble.com/Hash-entry-scavenge-pro=
blem.-tf3527308.html#a9842246
Sent from the JamVM mailing list archive at Nabble.com.