Shawn Walker wrote: > Roland Mainz wrote: > > Why was CR #6569725 > > (http://bugs.opensolaris.org/view_bug.do?bug_id=6569725 - "Add better > > large page out of box support for UltraSPARC I and II") marked as > > "11-Closed:Not a Defect (Closed)" ? > > UltraSPARC-I support was removed from Solaris 10, as per PSARC 2003/014 > EOL of 32-bit SPARC Kernels and CR 4847351 remove support for > UltraSPARC-I processors.
AFAIK that's a seperate issue since Rainer Orth and others want UltraSPARC-I support back in OS/Net... ... my issue is that "Large Page Support Out of the box" appears to do _nothing_ on my Ultra5 ([1]), Blade1000 or Ultra45. [1]=Someone can quickly argue that an Ultra5 doesn't have enougth memory to make proper use of LPOOTB but even in low-memory situations I would expect that the system at least uses 64K pages for process/thread stacks (this is usually area with many hotspots and it isn't very usefull to start paging here anyway). > Large Page Out of Box support was added to Nevada snapshot 52 for > UltraSPARC II, > as per the following CRs: > 6254029 memcntl() MC_HAT_ADVISE with page size 0 may cause segment page > sizes to be demoted > 6325885 map_pgszstk() uses p->p_brkpageszc rather than p->p_stkpageszc > 6371967 assign large pages to anon segment created using mmap /dev/zero > 6483208 unify and cleanup OOB (out of the box) large pagesize selection code > 6483216 use intermediate pagesizes to map the beginning of bss/heap and > stack when it may help performance > 6483226 bss size is not properly taken into account by LP OOB policy at > exec() time > 6483230 grow_internal() doesn't properly align stack bottom for large pages > 6483231 memcntl.c: ASSERT(IS_P2ALIGNED(p->p_brkbase + p->p_brksize, pgsz)); > 6485171 memcntl() shouldn't silently fail when stack space is > unavailable with requested pagesize Is there anything special I need to do to enable this support ? Somehow I'd expect the use of 64K pages if the stack grows beyond 64k in size on all UltraSPARC CPUs (even on US-3/4 since the number of stack pages easily fits in the 16slot MMU thing) and the use of 64K pages when heap/anon grows beyond 64K on UltraSPARC-2+Niagara1/2/2+ and the use of 4M pages if heap/anon/stack grows beyond 4M etc. - but nothing happens, even if my application allocates 8MB stack and 300MB heap. ---- Bye, Roland -- __ . . __ (o.\ \/ /.o) roland.mainz at nrubsig.org \__\/\/__/ MPEG specialist, C&&JAVA&&Sun&&Unix programmer /O /==\ O\ TEL +49 641 3992797 (;O/ \/ \O;)
