Kernel

SCO OpenServer Release 6.0.0 is based on the SCO System V Release 5 (SVR5) UNIX kernel. The new SCO OpenServer Release 6.0.0 kernel provides many enhancements and improvements over the original SCO OpenServer System V Release 3.2 kernel, as well as some differences.

Some of the highlights of the advanced SCO OpenServer Release 6.0.0 kernel include:

• Fully multi-threaded, with fine-grained locking, providing increased scalability and better real-time performance.

• Lightweight processes (LWPs) are supported, allowing parallel execution of multiple threads for the same process. See the _lwp* system calls in the Section (S) manual pages.

• The process scheduler allows timesharing and fixed-priority workloads.

• Dynamically loadable modules provide for the loading of device drivers without a kernel relink and reboot.

• The link_unix(ADM) command, by default, now defers kernel relinks until the system is rebooted.

• Kernel autotuning allows for some kernel tunables to be computed from the memory size detected at system startup.

• Each of the kernel data structures that keep track of users, groups, processes, and devices now use expanded type sizes. This allows for a greater number of system-wide UIDs, GIDs, PIDs, devices, and inodes.

• All SVR5 and SCO OpenServer device names are supported. SCO OpenServer device names take precedence if there is a clash.

• /dev/kmem notes:

  • Because of the differences between the SVR5 and SCO OpenServer Release 5 kernels, there are many changes in structure and symbol names. Some SCO OpenServer Release 5 applications that use nlist(ELF) to look up kernel symbols prior to reading/writing /dev/kmem may have problems. Because nlist cannot find symbols in loadable modules, getksym(S) is preferred.

  • To help track and diagnose the potential misuse of kernel memory by older programs compiled with the OSR5 ABI, the SCO OpenServer Release 6.0.0 kernel can prevent and/or log writes to /dev/kmem. Two new kmem tunable parameters are provided:

    
    OSRKMEMWDISABLE

    if set to ``1'' (default), the system disallows writes to /dev/kmem by programs compiled to the OSR5 ABI.

    
    OSRKMEMWLOG

    if set to ``1'' (default), the system logs a message to syslog each time a program that is compiled to the OSR5 ABI writes, or attempts to write, to /dev/kmem.

    To change either of these parameters, run:
    
    # cd /etc/conf/bin
    # ./idtune OSRKMEMWDISABLE [ 0 | 1 ]
    # ./idtune OSRKMEMWLOG [ 0 | 1 ]

    Then relink the kernel and reboot the system.

  • See the updated kmem(HW) manual page for details.

• The /dev/string and /dev/table directories are not included in SCO OpenServer Release 6.0.0. The SCO OpenServer Release 5 special files contained in these directories are incompatible with the new SCO OpenServer Release 6.0.0 kernel. Similar system information can be viewed using the resmgr(ADM) and dcu(ADM) commands.

• The /etc/inittab file is now sourced from /etc/conf/init.d/kernel instead of /etc/conf/cf.d/init.base. See the inittab(F), enable(C), and disable(C) manual pages.

• The trace(CP) system call tracer program has been removed from SCO OpenServer Release 6.0.0. Use the truss(C) command instead.

• The pstat(C) command has been removed. Equivalent information can be obtained from the crash(ADM) utility.

• The kprf(ADM) and kprpt(ADM) kernel profiling commands are no longer provided. New versions of prfstat(ADM), prfsnap(ADM), and prfpr(ADM), which provide similar functionality, are now available. See the profiler(ADM) manual page for more information.

• The eisa(ADM), pipe(ADM), pipestat(ADM), and tickadj(ADM) commands are now obsolete.

• The following XENIX emulation features aren't supported:

  • in-filesystem XENIX semaphores ( mknod -s option)

  • XENIX shared memory segments ( mknod -m option)

  • mandatory file-locking [see locking(S-osr5)]

  • device ioctl commands

  See the xemul(C) manual page for more information.

Linking the kernel

Enhancements in the SCO OpenServer Release 6.0.0 link kit have expanded the options for how you relink the kernel.

The /etc/conf/cf.d/link_unix script is retained and functions as it did on SCO OpenServer Release 5 systems, with one exception: when link_unix(ADM) is executed with no options, it does not prompt whether to use the kernel for the next reboot nor whether to rebuild the kernel environment. Instead, it executes idbuild(ADM) with no options, as described below.

When you use the /etc/conf/bin/idbuild command with no options to build the kernel, it records the fact that a kernel relink is necessary. A relink is automatically performed the next time the system is booted. This new behavior allows you to minimize the number of relinks that are run when adding kernel components or packages. See the idbuild(ADM) manual page for more information.

To avoid the need for a second reboot, you can run /etc/conf/cf.d/link_unix with the -y option. Note that all kernel configuration changes must be made before running link_unix -y. Be sure that you only run link_unix -y once and reboot the system immediately afterwards. Alternatively, you can use idbuild -B to rebuild the kernel immediately.

Dynamically Loadable Kernel Modules

Dynamically Loadable Kernel Modules (DLKMs) allow system resources to be conserved and allow you to make changes in the kernel configuration without the need to bring the system down and back up again. DLKMs are kernel modules that can be loaded automatically while the system is running. This allows you to keep the kernel smaller because unused drivers are not loaded into memory.

Examples of DLKMs are device drivers, filesystems, and STREAMS modules. Previously, to install and use a new module, you had to rebuild the kernel and reboot the system. Now, you can use a DLKM and avoid this work and the delay. Floppy disk drivers and mouse drivers are two examples of kernel modules that are typically configured as DLKMs.

Dynamically Loadable Kernel Modules can be autoloaded by the kernel, or demand-loaded or -unloaded by you. You should not need to demand-load modules. The system does not allow you to demand-unload a module unless it is no longer in use. Note that demand-unload is not a widely used function and should be used with some caution. It could expose rarely observed problems in the driver you unload. See ``Managing dynamically loadable kernel modules'' and the idbuild(ADM) manual page for more information.

Enabling the kernel debugger (KDB)

The SCODB kernel debugger is no longer part of the product. SCO OpenServer Release 6.0.0 provides the kdb(ADM) kernel debugger in its place.

kdb(ADM) is not enabled by default. To enable it:

1. Edit /etc/conf/sdevice.d/kdb and /etc/conf/sdevice.d/kdb_util and make the following change in both files:

   • change field 2 of the driver line from ``N'' to ``Y''

2. Run the following commands:
   
   # /etc/conf/bin/idbuild -M kdb_util
   # /etc/conf/bin/idbuild -M kdb
   # modadmin -l kdb_util
   # modadmin -l kdb

Once kdb is enabled, press <Ctrl><Alt>d to enter the debugger. Then you can run scodb to enter a SCODB-like command line interface.

Handling panic dumps

SCO OpenServer Release 6.0.0 is designed to save panic dumps if the system detects one is present upon reboot. By default, panic dumps are saved in /var/spool/dump. The dumps are saved as files using the following naming convention:

   panic.YYYY-MM-DD.hh:mm.bz2

where YYYY-MM-DD.hh:mm are the year, month, day, hour, and minute, respectively, of the panic dump.

Note that the files are compressed with bzip2 and thus use the .bz2 extension.

The manner in which panic dumps are saved can be customized in the /etc/default/dump file. Options include:

DUMP_DIRECTORY

/var/spool/dump by default

DUMP_COMPRESSOR

set to bzip2

DUMP_ENCRYPTER

no encryption if unset, or you could opt to use crypt

To analyze a saved panic dump, run the following commands:

# cd /var/spool/dump
# bunzip2 panic.YYYY-MM-DD.hh:mm.bz2
# /usr/sbin/crash -n /stand/unix -d panic.YYYY-MM-DD.hh:mm