Hi,

a couple of days ago, Maurizio pointed me to something that could be
described as a bunch of memory management regressions. These bugs can be
usually reproduced by simulatenously running two instances of tester
malloc1 or an instance of tester malloc1 and kconsole's test * on a SMP
system (e.g. -smp 4 in QEMU). I've started to hunt these bugs down and
already performed several tests to verify or disprove various theories.

I think there are two main categories of bugs:

- uspace and kernel data corruptions (possibly including tickets with
symptoms such uspace malloc heap corruption)

#396    tester malloc1 sometimes fails
#451    kernel panic when running tester malloc1/2 on 4-way SMP

- kernel deadlocks

#458    Deadlocks when memory management is under pressure

The data corruptions are reproducible on amd64 (maybe exclusively), the
deadlocks seem to be more universal.

I already found and fixed some issues in my lp:~jakub/helenos/mm branch,
but apparently not enough to close a single one of the tickets above.

Jakub

_______________________________________________
HelenOS-devel mailing list
[email protected]
http://lists.modry.cz/cgi-bin/listinfo/helenos-devel

Reply via email to