solaris.xtra.4095849

基于ntp协议的网络时间服务程序

 Bug Id: 4095849
 Category: kernel
 Subcategory: syscall
 State: evaluated
 Synopsis: time_constant value >6 with PLL in use leads to integer divide
           zero trap panic
 Description:
If the time_constant parameter is 7 or higher, and the phase-lock looping model
is in use, the system will take a "integer divide zero trap" panic in
the clock routine as soon as the time_offset becomes non-zero.

time_constant defaults to 0.  The only place it is set is in the ntp_adjtime
system call, from the 'constant' element of the timex structure argument.

 Work around:
Never set the constant element of the timex structure passed to ntp_adjtime to
a value larger than 6.

satish.mynam@Eng 1998-04-30
1. Use Sun's version of NTP software instead of PD version. This problem
is not seen with Sun's NTP version (which is mostly eqivalent to PD NTP 3.4
plus some Sun's local functionality futures).

2. Workaround for the public domain NTP version  ONLY:
 =====================================================
The workaround for public domain NTP version is to disable the
KERNEL_PLL from the NTP code. This way ntp_Adjtime() system call is
totally bypassed without sacrificing any of the functionality of the
NTP. The only hit you might see is the way  kernel precision timminig
is done without the PLL algorithm in the kernel.

    The easiest way to disable ntp_adjtime option is(without changing
    any makefiles or other config files) to disable the KERNEL_PLL
    value in the ./config.h file.

After doing a ./configure for probing for all the necessary tools(compilers,
os version, libraries),  please comment out KERNEL_PLL macro in
the ./config.h  file.  This will disable the KERNEL_PLL part of the source
code and the newly obtained xntpd is just similar to the old one but it
does not use ntp_adjtime() system call. This prevents it from panic'ng
the kernel.

/*#define KERNEL_PLL 1*/

I complied a new xntpd binary this way and it does nothave any ntp_adjtime()
related stuff.

Default:
=======
/net/divya/export/home/mynam/public_domain/ntp/xntp3-5.92/xntpd>strings
xntpd |
grep ntp_adjtime
354:adj_frequency: ntp_adjtime failed: %m
357:loop_config: ntp_adjtime() failed: %m
435:get_kernel_info: ntp_adjtime() failed: %m

With KERNEL_PLL disabled in config.h file
-=======================

/net/divya/export/home/mynam/public_domain/ntp/xntp3-5.92/xntpd>strings
xntpd.nopll | grep ntp_adjtime

        Integrated in releases:
 Duplicate of:
 Patch id:
 See also: 4133517
 Summary:
If the time_constant parameter is 7 or higher, and the phase-lock looping model
is in use, the system will take a "integer divide zero trap" panic in
the clock routine as soon as the time_offset becomes non-zero.

time_constant defaults to 0.  The only place it is set is in the ntp_adjtime
system call, from the 'constant' element of the timex structure argument.
----------------------------------------------------------------------------