Skip to contentSkip to conversation

2.6.14-rc5 mlock/munlock bug

DISCUSSION Mr. Berkley Shands 2 participants 2 messages Oct 21, 2005 – Oct 25, 2005

lore
Share & tools
lore
#0 lore raw

I've noticed that between 2.6.14-rc2 and 2.6.14-rc5 mlock() and mlunlock() have broken.

a call to mlock() to lock pages is granted, and the pages are locked, but never unlocked, even when munlock() is manually called or at process rundown.
No problems under 2.6.13 or up to 2.6.14-rc2. But sometime after -rc2 it goes BANG and the machine gets very unhappy. If you look at "swapon -s" you see more and more swap space is used until there is no physical memory left, then things really get unhappy.

berkley

#1 lore raw
Quoted 4 lines · Mr. Berkley Shands

I've noticed that between 2.6.14-rc2 and 2.6.14-rc5 mlock() and mlunlock() have broken.

Quoted 10 lines

a call to mlock() to lock pages is granted, and the pages are locked, but never
unlocked, even when munlock() is manually called or at process rundown. No problems under 2.6.13 or up to 2.6.14-rc2. But sometime after -rc2 it goes BANG
and the machine gets very unhappy. If you look at "swapon -s" you see more and more swap
space is used until there is no physical memory left, then things really get unhappy.

I cannot reproduce this behaviour here. Can you please provide a testcase?

This is on x86, I assume?