Difference between revisions of "Download/kernel/rhel6-testing/042stab072.2/changes"

From OpenVZ Virtuozzo Containers Wiki
Jump to: navigation, search
(I guess our memory scheduler is not enchanted. Yet.)
m (Kir moved page Download/kernel/rhel6-testing/042stab071.2/changes to Download/kernel/rhel6-testing/042stab072.2/changes without leaving a redirect: It's 072.2 now ))
 
(5 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
== Changes ==
 
== Changes ==
 +
<!-- 071.1 is a merge between 068.8 and 070.11 kernels, adding the features of 068.8 into 070.11 kernel -->
 
Since {{kernel link|rhel6-testing|042stab070.11}}:
 
Since {{kernel link|rhel6-testing|042stab070.11}}:
<!-- This is a merge between 068.x branch and 070.x branch. 071.1 kernel contains all patches from 068.8 and 070.12 kernels. -->
+
* Rebase to RHEL6.3 kernel 2.6.32-279.19.1.el6 (security and bug fixes, see link below)
 
<!-- 065.5 -->
 
<!-- 065.5 -->
 
<!-- 065.6 -->
 
<!-- 065.6 -->
 
<!-- 065.7 -->
 
<!-- 065.7 -->
 
<!-- 067.1 -->
 
<!-- 067.1 -->
* [ubc/mm] A new strict page cache isolation feature (PSBM-16955)
+
* [ubc/mm] A new strict page cache isolation feature (PSBM-16955, see [[Page cache isolation]])
** This feature disables bouncing pagecache pages between host and containers on physpages shortage in container. Internal reclaimer will drop cached data if container exceeded its physpages limit and pagecache isolation turned on.
 
** Current state: /proc/bc/<id>/debug:pagecache_isolation (disabled by default)
 
** sysctls:
 
*** ubc.pagecache_isolation = 0|1      turn off|on isolation for all containers
 
*** ubc.pagecache_isolation_on = <id>  turn on for container <id> (write only)
 
*** ubc.pagecache_isolation_off = <id>  turn off for container <id> (write only)  
 
 
<!-- 067.2 -->
 
<!-- 067.2 -->
 
* [mm] per-bc pagecache should be shrinked proportionally now (PSBM-15799)
 
* [mm] per-bc pagecache should be shrinked proportionally now (PSBM-15799)
Line 19: Line 14:
 
* [dcache] introduce per-bc dcache lru lists with real time shrink. Underlying dcache can be immediately shrunk if sysctl "ubc.dcache_no_vzfs_cache" is set (PSBM-16072)
 
* [dcache] introduce per-bc dcache lru lists with real time shrink. Underlying dcache can be immediately shrunk if sysctl "ubc.dcache_no_vzfs_cache" is set (PSBM-16072)
 
* [vSwap] virtual vSwap feature has been introduced => now vSwap does not require free physical swap on the HN (PSBM-16764)
 
* [vSwap] virtual vSwap feature has been introduced => now vSwap does not require free physical swap on the HN (PSBM-16764)
* [pagecache] enhanced version of memory scheduler: shadow gangs, reclaimer takes timestamps into the consideration, etc. (PSBM-15796)
+
* [pagecache] enchanted version of memory scheduler: shadow gangs, reclaimer takes timestamps into the consideration, etc. (PSBM-15796)
** vmscan priority is calculated for all LRU vectors according to its memory usage/limit and age of LRU lists.
 
** sysctls:
 
*** vm.usage_factor  (default 2)
 
*** vm.shadow_factor (default 8)
 
*** vm.age_factor    (default 8)
 
*** priority = 12 + usage * usage_factor / limit + shadow * shadow_factor / limit (for shadow part only) + age * age_factor / max_age
 
 
<!-- 068.3 -->
 
<!-- 068.3 -->
 
* [ubc/dcache] fix endless oom postponing in dcache reclaimer (PSBM-17067, PSBM-17069)  
 
* [ubc/dcache] fix endless oom postponing in dcache reclaimer (PSBM-17067, PSBM-17069)  
Line 36: Line 25:
 
<!-- 068.7 was a "side" kernel -->
 
<!-- 068.7 was a "side" kernel -->
 
<!-- 068.8 -->
 
<!-- 068.8 -->
* [devpts] handle dentry allocation failure properly in devpts_pty_new() ({{B|1758}})
 
 
<!-- 070.12 -->
 
<!-- 070.12 -->
 
* [fuse] a missed fuse_file_put() has been added
 
* [fuse] a missed fuse_file_put() has been added
 
* [mm] a panic while using huge tables has been fixed (PSBM-17315)
 
* [mm] a panic while using huge tables has been fixed (PSBM-17315)
 +
<!-- 071.2 -->
 +
* [media/rc] a possible kernel panic on "rc-core" module unload has been fixed
 +
* [fs/epoll] spurious lockdep warnings has been eliminated
 +
* [mm/hugepages] hugepages are now accounted inside a CT (to kmemsize and lockedpages) (PSBM-17327)
 +
* [ubc] swap failcounters should not be incremented in case of non fatal shortage
 +
* [mm] a new formula has been implemented for pagecache reclaimer (PSBM-15799)
 +
* [ubc] fix BUG due to sleep-in-atomic in set_gang_limits()
 +
* [mm] limit the CT memory portion by its commitment. In simple words: if you have single 256mb container on 1gb machine it's portion should be 256m, not 1gb.
 +
* [mm] set commitment to 0 for unlimited containers. For vmscan priorities that means that their effective memory portion is exactly equal to memory usage and give no bonuses for eliminating penalties for aged pages.
 +
* [cpt] make vzreboot be able to work with new vSwap implementation (without necessity of physical swap) (PSBM-17124)
 +
* [ploop] fix lockdep complains about bd_full acquiring (PSBM-13653)
 +
<!-- 072.1: rebase to 2.6.32-279.19.1.el6 -->
 +
 +
<!-- 072.2 -->
 +
* [mm] do not force scanning top priority lru vectors, this could result into forced push anon memory to swap (PSBM-17443)
 +
* [mm] struct gang size decreased to 16Kb, this should eliminate often memory allocation failures on CT start (PSBM-17433)
 +
* [fs] lockdep locking classes for virtualized filesystem types has been fixed
 +
* [fs] CT O_DIRECT and fsync settings checks have been corrected (PSBM-17157)
 +
* [pfcache] fix using uninitialized data on pfcache mounting
 +
 +
=== See also ===
 +
* {{RHSA|2012-1580}}

Latest revision as of 23:14, 24 December 2012

Changes[edit]

Since 042stab070.11:

  • Rebase to RHEL6.3 kernel 2.6.32-279.19.1.el6 (security and bug fixes, see link below)
  • [ubc/mm] A new strict page cache isolation feature (PSBM-16955, see Page cache isolation)
  • [mm] per-bc pagecache should be shrinked proportionally now (PSBM-15799)
  • [mm] debug trace events has been added into shrink_zone and shrink_gang
  • [dcache] introduce per-bc dcache lru lists with real time shrink. Underlying dcache can be immediately shrunk if sysctl "ubc.dcache_no_vzfs_cache" is set (PSBM-16072)
  • [vSwap] virtual vSwap feature has been introduced => now vSwap does not require free physical swap on the HN (PSBM-16764)
  • [pagecache] enchanted version of memory scheduler: shadow gangs, reclaimer takes timestamps into the consideration, etc. (PSBM-15796)
  • [ubc/dcache] fix endless oom postponing in dcache reclaimer (PSBM-17067, PSBM-17069)
  • [mm] the number of vmscan prios has been increased from 25 to 42. vmscan iterator has been fixed. (PSBM-15799)
  • [cpt] kernel panic has been fixed due to DEAD task restoration attempt (PCLIN-30483, PSBM-17107)
  • [mm] deadlock in kswapd near oom condition has been fixed (PSBM-17100)
  • [fuse] a missed fuse_file_put() has been added
  • [mm] a panic while using huge tables has been fixed (PSBM-17315)
  • [media/rc] a possible kernel panic on "rc-core" module unload has been fixed
  • [fs/epoll] spurious lockdep warnings has been eliminated
  • [mm/hugepages] hugepages are now accounted inside a CT (to kmemsize and lockedpages) (PSBM-17327)
  • [ubc] swap failcounters should not be incremented in case of non fatal shortage
  • [mm] a new formula has been implemented for pagecache reclaimer (PSBM-15799)
  • [ubc] fix BUG due to sleep-in-atomic in set_gang_limits()
  • [mm] limit the CT memory portion by its commitment. In simple words: if you have single 256mb container on 1gb machine it's portion should be 256m, not 1gb.
  • [mm] set commitment to 0 for unlimited containers. For vmscan priorities that means that their effective memory portion is exactly equal to memory usage and give no bonuses for eliminating penalties for aged pages.
  • [cpt] make vzreboot be able to work with new vSwap implementation (without necessity of physical swap) (PSBM-17124)
  • [ploop] fix lockdep complains about bd_full acquiring (PSBM-13653)
  • [mm] do not force scanning top priority lru vectors, this could result into forced push anon memory to swap (PSBM-17443)
  • [mm] struct gang size decreased to 16Kb, this should eliminate often memory allocation failures on CT start (PSBM-17433)
  • [fs] lockdep locking classes for virtualized filesystem types has been fixed
  • [fs] CT O_DIRECT and fsync settings checks have been corrected (PSBM-17157)
  • [pfcache] fix using uninitialized data on pfcache mounting

See also[edit]