Re: [Openmosix-devel] Re: [openMosix-general] Shared memory patch

[Moshe B. <mo...@mo...>]

why is a new syscall necessary and not a simple sysctl?

Moshe


On Wednesday, Mar 26, 2003, at 06:53 US/Central, MAASK Group wrote:

> Hi Moshe,
>
>> Since we are on this topic, we have to find another way around the 
>> two new syscalls. It's generally not a very good idea to add >> 
>> syscalls.
> --- We agree. sys_printlogs() is not necessary... we can use /proc 
> here. But the second system call... sys_flush() is necessary.
>
> Regards,
> MAASK
>>
>>
>> On Wednesday, Mar 26, 2003, at 06:29 US/Central, MAASK Group wrote:
>>
>>> Hi Dmitry,
>>>
>>>> I decided to try your patch, providing DSM capabilities.
>>>> That's really a cool thing!
>>> --- Hey, thanks!
>>>
>>>> I want to apply the patch to the latest CVS openMosix snapshot
>>>> with 2.4.20 kernel (while patch is 2.4.19 kernel oriented).
>>>> The patch applied more or less ok. Two main failures are
>>>> connected with entry.S and unistd.h.
>>> --- ya. We have added two system calls.
>>>
>>>> I have added new entries (according to patch) just to the end of
>>>> arch/i386/kernel/entry.S:
>>>>
>>>>     .long SYMBOL_NAME(sys_ni_syscall)   /* 250 sys_alloc_hugepages 
>>>> */
>>>>     .long SYMBOL_NAME(sys_ni_syscall)   /* sys_free_hugepages */
>>>>     .long SYMBOL_NAME(sys_ni_syscall)   /* sys_exit_group */
>>>> #ifdef CONFIG_SHM
>>>>     .long SYMBOL_NAME(sys_printlogs)
>>>>     .long SYMBOL_NAME(sys_flush)
>>>> #endif /* CONFIG_SHM */
>>>>
>>>>     .rept NR_syscalls-(.-sys_call_table)/4
>>>>         .long SYMBOL_NAME(sys_ni_syscall)
>>>>     .endr
>>>>
>>>> and into include/asm-i386/unistd.h:
>>>>
>>>> #define __NR_io_cancel      249
>>>> #define __NR_alloc_hugepages    250
>>>> #define __NR_free_hugepages 251
>>>> #define __NR_exit_group     252
>>>> #ifdef CONFIG_SHM
>>>> #define __NR_printlogs      253
>>>> #define __NR_flush          254
>>>> #endif /* CONFIG_SHM */
>>>>
>>>> At last I've got the next compilation (more exacly linkage) error:
>>>> pilot:/usr/src/linux-2.4.20-mosix # make bzImage > /dev/null
>>>> ...
>>>> arch/i386/kernel/kernel.o: In function `remote_sys_call_table':
>>>> arch/i386/kernel/kernel.o(.data+0x464): undefined reference to 
>>>> `remote_sys_printlogs'
>>>> make: *** [vmlinux] Error 1
>>>> That `remote_sys_printlogs' was found in arch/i386/kernel/mosasm.H:
>>>>
>>>> #ifdef CONFIG_SHM
>>>>     .long SYMBOL_NAME(remote_sys_printlogs)
>>>>     .long SYMBOL_NAME(remote_sys_flush)
>>>> #endif /* CONFIG_SHM */
>>>>
>>>> which is automatically generated in Makefile from entry.S.
>>>> Of course, that syscalls are not defined anywhere.
>>> --- sys_printlogs() is defined in linux/mig_shm/log_info.c
>>>     It just prints the shared memory structure defined in Migshm... 
>>> (shmeminfo)
>>>     It is basically there for debugging purposes and for the 
>>> developer to know whats going on with the shared memory regions on 
>>> this node...  Removing it from the kernel will not hamper the 
>>> current working.
>>>     But, this syscall is to be executed on the deputy... so 
>>> remote_sys_printlogs should not be generated.... check if you have 
>>> made an entry in hpc/syscalls.c by mistake.... if yes, remove it.
>>>    And finally, even if that doesnt work... just replace your..
>>>
>>>> #ifdef CONFIG_SHM
>>>>     .long SYMBOL_NAME(remote_sys_printlogs)
>>>>     .long SYMBOL_NAME(remote_sys_flush)
>>>> #endif /* CONFIG_SHM */
>>>
>>> by...
>>>
>>>> #ifdef CONFIG_SHM
>>>>     .long SYMBOL_NAME(sys_printlogs)
>>>>     .long SYMBOL_NAME(remote_sys_flush)
>>>> #endif /* CONFIG_SHM */
>>>
>>> .... and do a make bzImage only. Beleive me, it will work!
>>>
>>>> Finally, I don't know what to do. I can correct the program
>>>> doing that (offset.c) to skip this syscalls.
>>> --- umm... No. That is not the right thing to do I guess!
>>> But I don't know, how it will turn.
>>>>
>>>> Can anyone (kernel hacker) me here?
>>> ---- Do send in the results... and also the patch for 2.4.20 if it 
>>> works out :)
>>>
>>> Regards,
>>> Asmita from MAASK
>>>
>>> _________________________________________________________________
>>> Call US for just Rs. 5. http://www.msn.co.in/webtelephony/ Get a 
>>> phone card
>>>
>>>
>>>
>>> -------------------------------------------------------
>>> This SF.net email is sponsored by:
>>> The Definitive IT and Networking Event. Be There!
>>> NetWorld+Interop Las Vegas 2003 -- Register today!
>>> http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en
>>> _______________________________________________
>>> Openmosix-devel mailing list
>>> Ope...@li...
>>> https://lists.sourceforge.net/lists/listinfo/openmosix-devel
>>>
>>
>>
>>
>> -------------------------------------------------------
>> This SF.net email is sponsored by:
>> The Definitive IT and Networking Event. Be There!
>> NetWorld+Interop Las Vegas 2003 -- Register today!
>> http://ads.sourceforge.net/cgi-bin/redirect.pl?keyn0001en
>> _______________________________________________
>> Openmosix-devel mailing list
>> Ope...@li...
>> https://lists.sourceforge.net/lists/listinfo/openmosix-devel
>
>
> _________________________________________________________________
> Get ball by ball action on your desktop. 
> http://server1.msn.co.in/msnspecials/cricketdownload/contest.asp Get 
> Hutch MSN Cricketer
>