summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSamuel Thibault <samuel.thibault@ens-lyon.org>2025-02-04 20:58:25 +0100
committerSamuel Thibault <samuel.thibault@ens-lyon.org>2025-02-04 20:58:25 +0100
commit3f655a54b12710c460d591a98b5bdfb300913952 (patch)
tree7156cdd7dafdf083b756b6504f5fa55fffff3104
parentbeccce6240bb8db5981187eda7c2519b6092e1e7 (diff)
Notes about out of memory situations
-rw-r--r--open_issues/mach-defpager_swap.mdwn15
1 files changed, 14 insertions, 1 deletions
diff --git a/open_issues/mach-defpager_swap.mdwn b/open_issues/mach-defpager_swap.mdwn
index 6e4dc088..2363bb9f 100644
--- a/open_issues/mach-defpager_swap.mdwn
+++ b/open_issues/mach-defpager_swap.mdwn
@@ -1,4 +1,4 @@
-[[!meta copyright="Copyright © 2012 Free Software Foundation, Inc."]]
+[[!meta copyright="Copyright © 2012,2025 Free Software Foundation, Inc."]]
[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
@@ -13,6 +13,19 @@ License|/fdl]]."]]"""]]
[[!toc]]
+# Out of memory situations
+
+GNU Mach currently does not know how much room mach-defpager has. When it is
+full, mach-defpager will start not being able to write data and thus just
+lose it. The corresponding object will trigger SIGBUS on access to lost data
+(hopefully killing the corresponding process, thus releasing memory).
+
+GNU Mach should be aware of how much is left, to better manage paging out,
+possibly throttling some processes. Hopefully ext2fs will be able to flush out
+dirty data and release memory. With overcommit control we can then avoid one
+process filling the whole memory+swap with anonymous data. If several processes
+cooperate in sucking memory, we'd have to OOM some of them.
+
# IRC, OFTC, #debian-hurd, 2012-06-16
<lifeng> I allocated a 5GB partition as swap, but hurd only found 1GB