Oracle setprivgrp/大文本页错误
问题描述
在 HPUX 上的dmesg输出或syslog中发现下面的信息
Process pid num= does not have mem locking privileges
^ or a meaningful PID with PHKL_/
[PHKL_]
required for using large text pages in this version
See setprivgrp command
这是怎么回事?
配置信息
操作系统 HPUX
版本
硬件系统 HP
系列 H
解决方法
这个显示表示执行了一个要求使用大文本页的二进制文件（因为chatr +pi L 在二进制文件上运行）用ADB 可以关闭该信息
echo allow_superpage_text?W | adb w /stand/vmunix
安装了PHKL_/后就无须再用这种方式关闭（要求重新启动并且如果重建内核必须重复执行）
如果不应使用该特性那么最好用chatr()重新设置二进制文件请求应用缺省的k 页如果应当使用该特性则进程执行时所在的组应当被赋予MLOCK 权限例如如果这是一个Oracle进程Oracle 用户所在的组为dba
setprivgrp dba CHOWN MLOCK
为永久性实现这个目的应当修改/etc/privgroup (位于/sbin/initd/set_prvgrp: /usr/sbin/setprivgrp f /etc/privgroup)（如有边要创建该文件添加一行dba CHOWN MLOCK）但如果使用了Advanced或Online JFS则必须安装下面的VxFS补丁程序以避免在装载可导致进程挂起无法杀死的二进制文件过程中死锁:
PHKL_ (s) 或 PHKL_ (s)
背景资料:
处理器需要将虚拟地址映射到物理页地址上它们使用一个硬件TLB (翻译后援缓沖器)来缓沖最近使用的译文如果译文不在TLB中就会出现读取页目录和更新TLB的开销PA 处理器与较老的处理器相比TLB故障的代价更大但是它可以处理更大的页面尺寸所以您需要的TLB登录项可能更少
HPUX 内核不能要求装载具有大文本页的二进制文件因此它必须在exec()过程中读一次然后将其锁入RAM中只有在进程具有MLOCK 权限的情况下内核才能进行此项处理而这个权限是在已经用setprivgrp 赋予当前组的情况下获得的
TLB故障的减少和文本页故障的消除应当能够提高系统性能代价就是启动时间稍微有些长（启动过程中装载整个的二进制文件）损失可用于其它目的的RAM页（现在内存中驻留整个二进制文件而不只是最近使用过的页面）
following with all English text
Oracle setprivgrp/large text pages errors
Problem Description
On HPUX I see the following message in the dmesg output or syslog:
Process pid num= does not have mem locking privileges
^ or a meaningful PID with PHKL_/
[PHKL_]
required for using large text pages in this version
See setprivgrp command
What is happening?
Configuration Info
Operating System HPUX
Version
Hardware System HP
Series H
Solution
This message means a binary was exec()uted which asked to use large text pages (because chatr +pi L was run on the binary file)
The message can be disabled with ADB:
echo allow_superpage_text?W | adb w /stand/vmunix
Disabling it this way (which requires a reboot and must be repeated if the kernel was rebuilt) is no longer necessary with PHKL_/
If the feature should not be used using chatr() to reset the binary to request the default k pages is the preferred method If the
feature should be used the group the process is executing in should be granted the MLOCK privilege eg if this is an Oracle
process and the group the Oracle user is in is dba:
setprivgrp dba CHOWN MLOCK
To do this permanently /etc/privgroup (which is read in
/sbin/initd/set_prvgrp: /usr/sbin/setprivgrp f /etc/privgroup)
should be edited (create if necessary add a line dba CHOWN MLOCK)
However if Advanced or Online JFS is used the following VxFS patch
needs to be installed to prevent a deadlock during the load of the binary that would result in a hung unkillable process:
PHKL_ (s) or PHKL_ (s)
Background:
Processors need to map virtual to physical page addresses They use a
piece of hardware the TLB (Transaction Lookaside Buffer) to cache recently used translations If a translation is not in the TLB
overhead to read the page directory and update the TLB is incurred
PA processors have a larger penalty for a TLB miss than older processors but can handle larger page sizes so that you may need
fewer TLB entries
The HPUX kernel cannot demand load a binary with large text pages so it has to read it once during exec() and lock it into RAM
The kernel will only do that if the process has the MLOCK privilegewhich is granted if the current group has been given this privilege with setprivgrp
The reduction of TLB misses and elimination of text page faults should increase performance The cost is a slightly larger startup time (during which the whole binary is loaded) and the loss of RAM pages available
for other purposes (the whole binary is now resident in memory not just the recently used pages)
