Hi,

> I see that you checked the changes in. I'm ambivalent about the whole
> thing, but I guess this is the least sucky option for now. Although
> there is actually an advantage to using mmap: at least it gives *some*
> control over the placement of memory, whereas with malloc you're just
> assuming that it is laid out in the traditional manner.

Yes, but the previous way didn't make use of it (AFAICT).  At least,
when increasing the data limit, it mapped the modules at addresses
beyond the 2GB limit, which made it fail.

> Speaking of layout.. have you tried defining __USE_TOPDOWN_VM in
> amd64/include/vmparam.h and checking its effect? Do you still get
> the sbrk kernel messages with that enabled?

Hmm, no, didn't try that yet.  Will do so, but probably no earlier
than the weekend.

Ciao,
Wolfgang
-- 
ws@TooLs.DE                            Wolfgang Solfrank, TooLs GmbH