|| ||Andrea Arcangeli <email@example.com>|
|| ||Sat, 3 Apr 2004 03:54:27 +0200|
this avoids (harmless) warnings while freeing non-compound pages, adds
some more robustness check to the freelist by making sure no page has
PG_compound set, and most important it adds merging for filebacked
mappings and anonymous mappings in mprotect. I'm running it on my main
desktop and there are no problems so far.
I believe the VM side is feature complete, in order:
# -aa VM (remove rmap)
Two pending thing to clarify are swapsuspend/swapresume (works partially
for me because aic7xxx isn't swap-resume capable), and Christoph's xfs
oops on ppc that looks a miscompilation of some sort.
Changelog diff between 2.6.5-rc3-aa2 and 2.6.5-rc3-aa3:
Files 2.6.5-rc3-aa2/extraversion and 2.6.5-rc3-aa3/extraversion differ
Files 2.6.5-rc3-aa2/gfp-no-compound and 2.6.5-rc3-aa3/gfp-no-compound differ
Avoid flood of warnings while freeing a non-compound multipage.
Added bugchecks to be sure all pages in the freelist have PG_compound
unset (for robusteness, the check for PG_compound is zerocost anyways
since we check it in a bitmask).
Christoph's troubles with ppc seems to be a gcc miscompilation if it's
really the second bad_page in destroy_compound_page triggering.
The swap suspend/swap resume works for me w/o apparent VM issues,
aic7xxx simply hangs at resume, but Pavel said it's expected. I'll
later try to load it on my laptop and see if I've more luck with it.
Only in 2.6.5-rc3-aa3: mprotect-vma-merging
Resurrected mprotect anon-vma and file-backed merging. The file-backed
merging is IMHO the most interesting one, and it's for the first time
available in linux.
Files 2.6.5-rc3-aa2/prio-tree.gz and 2.6.5-rc3-aa3/prio-tree.gz differ
Added s390 compilation fix.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to firstname.lastname@example.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/